Artificial neural network

From Wikipedia, the free encyclopedia

Jump to navigationJump to search
Machine learning and
data mining
Kernel Machine.svg
Problems[show]
Supervised learning
(classification � regression)
[show]
Clustering[show]
Dimensionality reduction[show]
Structured prediction[show]
Anomaly detection[show]
Artificial neural networks[show]
Reinforcement learning[show]
Theory[show]
Machine-learning venues[show]
Glossary of artificial intelligence[show]
Related articles[show]
Portal-puzzle.svg Machine learning portal
vte

An artificial neural network is an interconnected group of nodes, inspired by a

simplification of neurons in a brain. Here, each circular node represents an
artificial neuron and an arrow represents a connection from the output of one
artificial neuron to the input of another.
Artificial neural networks (ANN) or connectionist systems are computing systems
that are inspired by, but not necessarily identical to, the biological neural
networks that constitute animal brains. Such systems "learn" to perform tasks by
considering examples, generally without being programmed with any task-specific
rules. For example, in image recognition, they might learn to identify images that
contain cats by analyzing example images that have been manually labeled as "cat"
or "no cat" and using the results to identify cats in other images. They do this
without any prior knowledge about cats, for example, that they have fur, tails,
whiskers and cat-like faces. Instead, they automatically generate identifying
characteristics from the learning material that they process.

An ANN is based on a collection of connected units or nodes called artificial

neurons, which loosely model the neurons in a biological brain. Each connection,
like the synapses in a biological brain, can transmit a signal from one artificial
neuron to another. An artificial neuron that receives a signal can process it and
then signal additional artificial neurons connected to it.

In common ANN implementations, the signal at a connection between artificial

neurons is a real number, and the output of each artificial neuron is computed by
some non-linear function of the sum of its inputs. The connections between
artificial neurons are called 'edges'. Artificial neurons and edges typically have
a weight that adjusts as learning proceeds. The weight increases or decreases the
strength of the signal at a connection. Artificial neurons may have a threshold
such that the signal is only sent if the aggregate signal crosses that threshold.
Typically, artificial neurons are aggregated into layers. Different layers may
perform different kinds of transformations on their inputs. Signals travel from the
first layer (the input layer), to the last layer (the output layer), possibly after
traversing the layers multiple times.

The original goal of the ANN approach was to solve problems in the same way that a
human brain would. However, over time, attention moved to performing specific
tasks, leading to deviations from biology. Artificial neural networks have been
used on a variety of tasks, including computer vision, speech recognition, machine
translation, social network filtering, playing board and video games and medical
diagnosis.

Contents
1 History
1.1 Hebbian learning
1.2 Backpropagation
1.3 Hardware-based designs
1.4 Contests
1.5 Convolutional neural networks
2 Models
2.1 Components of an artificial neural network
2.2 Neural networks as functions
2.3 Learning
2.4 Learning paradigms
2.5 Learning algorithms
3 Optimization
3.1 Algorithm
4 Algorithm in code
4.1 Phase 1: propagation
4.2 Phase 2: weight update
4.3 Pseudocode
5 Extension
5.1 Adaptive learning rate
5.2 Inertia
6 Modes of learning
7 Variants
7.1 Group method of data handling
7.2 Convolutional neural networks
7.3 Long short-term memory
7.4 Deep reservoir computing
7.5 Deep belief networks
7.6 Large memory storage and retrieval neural networks
7.7 Stacked (de-noising) auto-encoders
7.8 Deep stacking networks
7.9 Tensor deep stacking networks
7.10 Spike-and-slab RBMs
7.11 Compound hierarchical-deep models
7.12 Deep predictive coding networks
7.13 Networks with separate memory structures
7.14 Multilayer kernel machine
8 Neural architecture search
9 Use
10 Applications
10.1 Types of models
11 Theoretical properties
11.1 Computational power
11.2 Capacity
11.3 Convergence
11.4 Generalization and statistics
12 Criticism
12.1 Training issues
12.2 Theoretical issues
12.3 Hardware issues
12.4 Practical counterexamples to criticisms
12.5 Hybrid approaches
13 Types
14 Gallery
15 See also
16 References
17 Bibliography
18 External links
History
Warren McCulloch and Walter Pitts[1] (1943) created a computational model for
neural networks based on mathematics and algorithms called threshold logic. This
model paved the way for neural network research to split into two approaches. One
approach focused on biological processes in the brain while the other focused on
the application of neural networks to artificial intelligence. This work led to
work on nerve networks and their link to finite automata.[2]

Hebbian learning
In the late 1940s, D. O. Hebb[3] created a learning hypothesis based on the
mechanism of neural plasticity that became known as Hebbian learning. Hebbian
learning is unsupervised learning. This evolved into models for long-term
potentiation. Researchers started applying these ideas to computational models in
1948 with Turing's B-type machines. Farley and Clark[4] (1954) first used
computational machines, then called "calculators", to simulate a Hebbian network.
Other neural network computational machines were created by Rochester, Holland,
Habit and Duda (1956).[5] Rosenblatt[6] (1958) created the perceptron, an algorithm
for pattern recognition. With mathematical notation, Rosenblatt described circuitry
not in the basic perceptron, such as the exclusive-or circuit that could not be
processed by neural networks at the time.[7] In 1959, a biological model proposed
by Nobel laureates Hubel and Wiesel was based on their discovery of two types of
cells in the primary visual cortex: simple cells and complex cells.[8] The first
functional networks with many layers were published by Ivakhnenko and Lapa in 1965,
becoming the Group Method of Data Handling.[9][10][11]

Neural network research stagnated after machine learning research by Minsky and
Papert (1969),[12] who discovered two key issues with the computational machines
that processed neural networks. The first was that basic perceptrons were incapable
of processing the exclusive-or circuit. The second was that computers didn't have
enough processing power to effectively handle the work required by large neural
networks. Neural network research slowed until computers achieved far greater
processing power. Much of artificial intelligence had focused on high-level
(symbolic) models that are processed by using algorithms, characterized for example
by expert systems with knowledge embodied in if-then rules, until in the late 1980s
research expanded to low-level (sub-symbolic) machine learning, characterized by
knowledge embodied in the parameters of a cognitive model.[citation needed]

Backpropagation
A key trigger for renewed interest in neural networks and learning was Werbos's
(1975) backpropagation algorithm that made the training of multi-layer networks
feasible and efficient. Backpropagation distributed the error term back up through
the layers, by modifying the weights at each node.[7]

In the mid-1980s, parallel distributed processing became popular under the name
connectionism. Rumelhart and McClelland (1986) described the use of connectionism
to simulate neural processes.[13]

Support vector machines and other, much simpler methods such as linear classifiers
gradually overtook neural networks in machine learning popularity. However, using
neural networks transformed some domains, such as the prediction of protein
structures.[14][15]

In 1992, max-pooling was introduced to help with least shift invariance and
tolerance to deformation to aid in 3D object recognition.[16][17][18] In 2010,
Backpropagation training through max-pooling was accelerated by GPUs and shown to
perform better than other pooling variants.[19]

The vanishing gradient problem affects many-layered feedforward networks that used
backpropagation and also recurrent neural networks (RNNs).[20][21] As errors
propagate from layer to layer, they shrink exponentially with the number of layers,
impeding the tuning of neuron weights that is based on those errors, particularly
affecting deep networks.

To overcome this problem, Schmidhuber adopted a multi-level hierarchy of networks

(1992) pre-trained one level at a time by unsupervised learning and fine-tuned by
backpropagation.[22] Behnke (2003) relied only on the sign of the gradient (Rprop)
[23] on problems such as image reconstruction and face localization.

Hinton et al. (2006) proposed learning a high-level representation using successive

layers of binary or real-valued latent variables with a restricted Boltzmann
machine[24] to model each layer. Once sufficiently many layers have been learned,
the deep architecture may be used as a generative model by reproducing the data
when sampling down the model (an "ancestral pass") from the top level feature
activations.[25][26] In 2012, Ng and Dean created a network that learned to
recognize higher-level concepts, such as cats, only from watching unlabeled images
taken from YouTube videos.[27]

Earlier challenges in training deep neural networks were successfully addressed

with methods such as unsupervised pre-training, while available computing power
increased through the use of GPUs and distributed computing. Neural networks were
deployed on a large scale, particularly in image and visual recognition problems.
This became known as "deep learning".[citation needed]

Hardware-based designs
Computational devices were created in CMOS, for both biophysical simulation and
neuromorphic computing. Nanodevices[28] for very large scale principal components
analyses and convolution may create a new class of neural computing because they
are fundamentally analog rather than digital (even though the first implementations
may use digital devices).[29] Ciresan and colleagues (2010)[30] in Schmidhuber's
group showed that despite the vanishing gradient problem, GPUs make back-
propagation feasible for many-layered feedforward neural networks.

Contests
Between 2009 and 2012, recurrent neural networks and deep feedforward neural
networks developed in Schmidhuber's research group won eight international
competitions in pattern recognition and machine learning.[31][32] For example, the
bi-directional and multi-dimensional long short-term memory (LSTM)[33][34][35][36]
of Graves et al. won three competitions in connected handwriting recognition at the
2009 International Conference on Document Analysis and Recognition (ICDAR), without
any prior knowledge about the three languages to be learned.[35][34]

Ciresan and colleagues won pattern recognition contests, including the IJCNN 2011
Traffic Sign Recognition Competition,[37] the ISBI 2012 Segmentation of Neuronal
Structures in Electron Microscopy Stacks challenge[38] and others. Their neural
networks were the first pattern recognizers to achieve human-competitive or even
superhuman performance[39] on benchmarks such as traffic sign recognition (IJCNN
2012), or the MNIST handwritten digits problem.

Researchers demonstrated (2010) that deep neural networks interfaced to a hidden

Markov model with context-dependent states that define the neural network output
layer can drastically reduce errors in large-vocabulary speech recognition tasks
such as voice search.
GPU-based implementations[40] of this approach won many pattern recognition
contests, including the IJCNN 2011 Traffic Sign Recognition Competition,[37] the
ISBI 2012 Segmentation of neuronal structures in EM stacks challenge,[38] the
ImageNet Competition[41] and others.

Deep, highly nonlinear neural architectures similar to the neocognitron[42] and the
"standard architecture of vision",[43] inspired by simple and complex cells, were
pre-trained by unsupervised methods by Hinton.[44][25] A team from his lab won a
2012 contest sponsored by Merck to design software to help find molecules that
might identify new drugs.[45]

Convolutional neural networks

As of 2011, the state of the art in deep learning feedforward networks alternated
between convolutional layers and max-pooling layers,[40][46] topped by several
fully or sparsely connected layers followed by a final classification layer.
Learning is usually done without unsupervised pre-training. In the convolutional
layer, there are filters that are convolved with the input. Each filter is
equivalent to a weights vector that has to be trained.

Such supervised deep learning methods were the first to achieve human-competitive
performance on certain practical applications.[39]

Artificial neural networks were able to guarantee shift invariance to deal with
small and large natural objects in large cluttered scenes, only when invariance
extended beyond shift, to all ANN-learned concepts, such as location, type (object
class label), scale, lighting and others. This was realized in Developmental
Networks (DNs)[47] whose embodiments are Where-What Networks, WWN-1 (2008)[48]
through WWN-7 (2013).[49]

Models

This section may be confusing or unclear to readers. Please help us clarify the
section. There might be a discussion about this on the talk page. (April 2017)
(Learn how and when to remove this template message)

Neuron and myelinated axon, with signal flow from inputs at dendrites to outputs at
axon terminals
An artificial neural network is a network of simple elements called artificial
neurons, which receive input, change their internal state (activation) according to
that input, and produce output depending on the input and activation.

An artificial neuron mimics the working of a biophysical neuron with inputs and
outputs, but is not a biological neuron model.

The network forms by connecting the output of certain neurons to the input of other
neurons forming a directed, weighted graph. The weights as well as the functions
that compute the activation can be modified by a process called learning which is
governed by a learning rule.[50]

Components of an artificial neural network

Neurons
A neuron with label {\displaystyle j} j receiving an input {\displaystyle p_{j}(t)}
{\displaystyle p_{j}(t)} from predecessor neurons consists of the following
components:[50]

an activation {\displaystyle a_{j}(t)} {\displaystyle a_{j}(t)}, the neuron's

state, depending on a discrete time parameter,
possibly a threshold {\displaystyle \theta _{j}} \theta _{j}, which stays fixed
unless changed by a learning function,
an activation function {\displaystyle f} f that computes the new activation at a
given time {\displaystyle t+1} t+1 from {\displaystyle a_{j}(t)} {\displaystyle
a_{j}(t)}, {\displaystyle \theta _{j}} \theta _{j} and the net input {\displaystyle
p_{j}(t)} {\displaystyle p_{j}(t)} giving rise to the relation
{\displaystyle a_{j}(t+1)=f(a_{j}(t),p_{j}(t),\theta _{j})} {\displaystyle a_{j}
(t+1)=f(a_{j}(t),p_{j}(t),\theta _{j})},
and an output function {\displaystyle f_{out}} {\displaystyle f_{out}} computing
the output from the activation
{\displaystyle o_{j}(t)=f_{out}(a_{j}(t))} {\displaystyle o_{j}(t)=f_{out}(a_{j}
(t))}.
Often the output function is simply the Identity function.

An input neuron has no predecessor but serves as input interface for the whole
network. Similarly an output neuron has no successor and thus serves as output
interface of the whole network.

Connections, weights and biases

The network consists of connections, each connection transferring the output of a
neuron {\displaystyle i} i to the input of a neuron {\displaystyle j} j. In this
sense {\displaystyle i} i is the predecessor of {\displaystyle j} j and
{\displaystyle j} j is the successor of {\displaystyle i} i. Each connection is
assigned a weight {\displaystyle w_{ij}} w_{ij}.[50] Sometimes a bias term is added
to the total weighted sum of inputs to serve as a threshold to shift the activation
function.[51]

Propagation function
The propagation function computes the input {\displaystyle p_{j}(t)} {\displaystyle
p_{j}(t)} to the neuron {\displaystyle j} j from the outputs {\displaystyle o_{i}
(t)} {\displaystyle o_{i}(t)} of predecessor neurons and typically has the form[50]

{\displaystyle p_{j}(t)=\sum _{i}o_{i}(t)w_{ij}} {\displaystyle p_{j}(t)=\sum

_{i}o_{i}(t)w_{ij}}.
When a bias value is added with the function, the above form changes to the
following:[52]

{\displaystyle p_{j}(t)=\sum _{i}o_{i}(t)w_{ij}+w_{0j}} {\displaystyle p_{j}

(t)=\sum _{i}o_{i}(t)w_{ij}+w_{0j}} , where {\displaystyle w_{0j}} {\displaystyle
w_{0j}} is a bias.
Learning rule
The learning rule is a rule or an algorithm which modifies the parameters of the
neural network, in order for a given input to the network to produce a favored
output. This learning process typically amounts to modifying the weights and
thresholds of the variables within the network.[50]

Neural networks as functions

See also: Graphical models
Neural network models can be viewed as simple mathematical models defining a
function {\displaystyle \textstyle f:X\rightarrow Y} {\displaystyle \textstyle
f:X\rightarrow Y} or a distribution over {\displaystyle \textstyle X} \textstyle X
or both {\displaystyle \textstyle X} \textstyle X and {\displaystyle \textstyle
Y} \textstyle Y. Sometimes models are intimately associated with a particular
learning rule. A common use of the phrase "ANN model" is really the definition of a
class of such functions (where members of the class are obtained by varying
parameters, connection weights, or specifics of the architecture such as the number
of neurons or their connectivity).

Mathematically, a neuron's network function {\displaystyle \textstyle f(x)}

\textstyle f(x) is defined as a composition of other functions {\displaystyle
\textstyle g_{i}(x)} \textstyle g_{i}(x), that can further be decomposed into other
functions. This can be conveniently represented as a network structure, with arrows
depicting the dependencies between functions. A widely used type of composition is
the nonlinear weighted sum, where {\displaystyle \textstyle f(x)=K\left(\sum
_{i}w_{i}g_{i}(x)\right)} \textstyle f(x)=K\left(\sum _{i}w_{i}g_{i}(x)\right),
where {\displaystyle \textstyle K} \textstyle K (commonly referred to as the
activation function[53]) is some predefined function, such as the hyperbolic
tangent, sigmoid function, softmax function, or rectifier function. The important
characteristic of the activation function is that it provides a smooth transition
as input values change, i.e. a small change in input produces a small change in
output. The following refers to a collection of functions {\displaystyle \textstyle
g_{i}} \textstyle g_{i} as a vector {\displaystyle \textstyle g=(g_{1},g_{2},\ldots
,g_{n})} \textstyle g=(g_{1},g_{2},\ldots ,g_{n}).

ANN dependency graph

This figure depicts such a decomposition of {\displaystyle \textstyle f} \textstyle
f, with dependencies between variables indicated by arrows. These can be
interpreted in two ways.

The first view is the functional view: the input {\displaystyle \textstyle x}
\textstyle x is transformed into a 3-dimensional vector {\displaystyle \textstyle
h} \textstyle h, which is then transformed into a 2-dimensional vector
{\displaystyle \textstyle g} \textstyle g, which is finally transformed into
{\displaystyle \textstyle f} \textstyle f. This view is most commonly encountered
in the context of optimization.

The second view is the probabilistic view: the random variable {\displaystyle
\textstyle F=f(G)} \textstyle F=f(G) depends upon the random variable
{\displaystyle \textstyle G=g(H)} \textstyle G=g(H), which depends upon
{\displaystyle \textstyle H=h(X)} \textstyle H=h(X), which depends upon the random
variable {\displaystyle \textstyle X} \textstyle X. This view is most commonly
encountered in the context of graphical models.

The two views are largely equivalent. In either case, for this particular
architecture, the components of individual layers are independent of each other
(e.g., the components of {\displaystyle \textstyle g} \textstyle g are independent
of each other given their input {\displaystyle \textstyle h} \textstyle h). This
naturally enables a degree of parallelism in the implementation.

Two separate depictions of the recurrent ANN dependency graph

Networks such as the previous one are commonly called feedforward, because their
graph is a directed acyclic graph. Networks with cycles are commonly called
recurrent. Such networks are commonly depicted in the manner shown at the top of
the figure, where {\displaystyle \textstyle f} \textstyle f is shown as being
dependent upon itself. However, an implied temporal dependence is not shown.

Learning
See also: Mathematical optimization, Estimation theory, and Machine learning
The possibility of learning has attracted the most interest in neural networks.
Given a specific task to solve, and a class of functions {\displaystyle \textstyle
F} \textstyle F, learning means using a set of observations to find
{\displaystyle \textstyle f^{*}\in F} \textstyle f^{*}\in F which solves the task
in some optimal sense.

This entails defining a cost function {\displaystyle \textstyle C:F\rightarrow

\mathbb {R} } \textstyle C:F\rightarrow \mathbb {R} such that, for the optimal
solution {\displaystyle \textstyle f^{*}} \textstyle f^{*}, {\displaystyle
\textstyle C(f^{*})\leq C(f)} \textstyle C(f^{*})\leq C(f) {\displaystyle
\textstyle \forall f\in F} \textstyle \forall f\in F � i.e., no solution has a cost
less than the cost of the optimal solution (see mathematical optimization).

The cost function {\displaystyle \textstyle C} \textstyle C is an important concept

in learning, as it is a measure of how far away a particular solution is from an
optimal solution to the problem to be solved. Learning algorithms search through
the solution space to find a function that has the smallest possible cost.

For applications where the solution is data dependent, the cost must necessarily be
a function of the observations, otherwise the model would not relate to the data.
It is frequently defined as a statistic to which only approximations can be made.
As a simple example, consider the problem of finding the model {\displaystyle
\textstyle f} \textstyle f, which minimizes {\displaystyle \textstyle
C=E\left[(f(x)-y)^{2}\right]} \textstyle C=E\left[(f(x)-y)^{2}\right], for data
pairs {\displaystyle \textstyle (x,y)} \textstyle (x,y) drawn from some
distribution {\displaystyle \textstyle {\mathcal {D}}} \textstyle {\mathcal {D}}.
In practical situations we would only have {\displaystyle \textstyle N} \textstyle
N samples from {\displaystyle \textstyle {\mathcal {D}}} \textstyle {\mathcal {D}}
and thus, for the above example, we would only minimize {\displaystyle \textstyle
{\hat {C}}={\frac {1}{N}}\sum _{i=1}^{N}(f(x_{i})-y_{i})^{2}} \textstyle {\hat
{C}}={\frac {1}{N}}\sum _{i=1}^{N}(f(x_{i})-y_{i})^{2}. Thus, the cost is minimized
over a sample of the data rather than the entire distribution.

When {\displaystyle \textstyle N\rightarrow \infty } \textstyle N\rightarrow \infty

some form of online machine learning must be used, where the cost is reduced as
each new example is seen. While online machine learning is often used when
{\displaystyle \textstyle {\mathcal {D}}} \textstyle {\mathcal {D}} is fixed, it is
most useful in the case where the distribution changes slowly over time. In neural
network methods, some form of online machine learning is frequently used for finite
datasets.

Choosing a cost function

While it is possible to define an ad hoc cost function, frequently a particular
cost function is used, either because it has desirable properties (such as
convexity) or because it arises naturally from a particular formulation of the
problem (e.g., in a probabilistic formulation the posterior probability of the
model can be used as an inverse cost). Ultimately, the cost function depends on the
task.

Backpropagation
Main article: Backpropagation
A DNN can be discriminatively trained with the standard backpropagation algorithm.
Backpropagation is a method to calculate the gradient of the loss function
(produces the cost associated with a given state) with respect to the weights in an
ANN.

The basics of continuous backpropagation[9][54][55][56] were derived in the context

of control theory by Kelley[57] in 1960 and by Bryson in 1961,[58] using principles
of dynamic programming. In 1962, Dreyfus published a simpler derivation based only
on the chain rule.[59] Bryson and Ho described it as a multi-stage dynamic system
optimization method in 1969.[60][61] In 1970, Linnainmaa finally published the
general method for automatic differentiation (AD) of discrete connected networks of
nested differentiable functions.[62][63] This corresponds to the modern version of
backpropagation which is efficient even when the networks are sparse.[9][54][64]
[65] In 1973, Dreyfus used backpropagation to adapt parameters of controllers in
proportion to error gradients.[66] In 1974, Werbos mentioned the possibility of
applying this principle to Artificial neural networks,[67] and in 1982, he applied
Linnainmaa's AD method to neural networks in the way that is widely used today.[54]
[68] In 1986, Rumelhart, Hinton and Williams noted that this method can generate
useful internal representations of incoming data in hidden layers of neural
networks.[69] In 1993, Wan was the first[9] to win an international pattern
recognition contest through backpropagation.[70]

The weight updates of backpropagation can be done via stochastic gradient descent
using the following equation:

{\displaystyle w_{ij}(t+1)=w_{ij}(t)-\eta {\frac {\partial C}{\partial w_{ij}}}+\xi

(t)} {\displaystyle w_{ij}(t+1)=w_{ij}(t)-\eta {\frac {\partial C}{\partial
w_{ij}}}+\xi (t)}
where, {\displaystyle \eta } \eta is the learning rate, {\displaystyle C} C is the
cost (loss) function and {\displaystyle \xi (t)} \xi (t) a stochastic term. The
choice of the cost function depends on factors such as the learning type
(supervised, unsupervised, reinforcement, etc.) and the activation function. For
example, when performing supervised learning on a multiclass classification
problem, common choices for the activation function and cost function are the
softmax function and cross entropy function, respectively. The softmax function is
defined as {\displaystyle p_{j}={\frac {\exp(x_{j})}{\sum _{k}\exp(x_{k})}}}
p_{j}={\frac {\exp(x_{j})}{\sum _{k}\exp(x_{k})}} where {\displaystyle p_{j}} p_{j}
represents the class probability (output of the unit {\displaystyle j} j) and
{\displaystyle x_{j}} x_{j} and {\displaystyle x_{k}} x_{k} represent the total
input to units {\displaystyle j} j and {\displaystyle k} k of the same level
respectively. Cross entropy is defined as {\displaystyle C=-\sum
_{j}d_{j}\log(p_{j})} C=-\sum _{j}d_{j}\log(p_{j}) where {\displaystyle d_{j}}
d_{j} represents the target probability for output unit {\displaystyle j} j and
{\displaystyle p_{j}} p_{j} is the probability output for {\displaystyle j} j after
applying the activation function.[71]

These can be used to output object bounding boxes in the form of a binary mask.
They are also used for multi-scale regression to increase localization precision.
DNN-based regression can learn features that capture geometric information in
addition to serving as a good classifier. They remove the requirement to explicitly
model parts and their relations. This helps to broaden the variety of objects that
can be learned. The model consists of multiple layers, each of which has a
rectified linear unit as its activation function for non-linear transformation.
Some layers are convolutional, while others are fully connected. Every
convolutional layer has an additional max pooling. The network is trained to
minimize L2 error for predicting the mask ranging over the entire training set
containing bounding boxes represented as masks.

Alternatives to backpropagation include Extreme Learning Machines,[72] "No-prop"

networks,[73] training without backtracking,[74] "weightless" networks,[75][76] and
non-connectionist neural networks.

Learning paradigms
The three major learning paradigms each correspond to a particular learning task.
These are supervised learning, unsupervised learning and reinforcement learning.

Supervised learning
Supervised learning uses a set of example pairs {\displaystyle (x,y),x\in X,y\in Y}
(x, y), x \in X, y \in Y and the aim is to find a function {\displaystyle
f:X\rightarrow Y} f : X \rightarrow Y in the allowed class of functions that
matches the examples. In other words, we wish to infer the mapping implied by the
data; the cost function is related to the mismatch between our mapping and the data
and it implicitly contains prior knowledge about the problem domain.[77]

A commonly used cost is the mean-squared error, which tries to minimize the average
squared error between the network's output, {\displaystyle f(x)} f(x), and the
target value {\displaystyle y} y over all the example pairs. Minimizing this cost
using gradient descent for the class of neural networks called multilayer
perceptrons (MLP), produces the backpropagation algorithm for training neural
networks.

Tasks that fall within the paradigm of supervised learning are pattern recognition
(also known as classification) and regression (also known as function
approximation). The supervised learning paradigm is also applicable to sequential
data (e.g., for hand writing, speech and gesture recognition). This can be thought
of as learning with a "teacher", in the form of a function that provides continuous
feedback on the quality of solutions obtained thus far.

Unsupervised learning
In unsupervised learning, some data {\displaystyle \textstyle x} \textstyle x is
given and the cost function to be minimized, that can be any function of the data
{\displaystyle \textstyle x} \textstyle x and the network's output,
{\displaystyle \textstyle f} \textstyle f.

The cost function is dependent on the task (the model domain) and any a priori
assumptions (the implicit properties of the model, its parameters and the observed
variables).

As a trivial example, consider the model {\displaystyle \textstyle f(x)=a}

\textstyle f(x)=a where {\displaystyle \textstyle a} \textstyle a is a constant and
the cost {\displaystyle \textstyle C=E[(x-f(x))^{2}]} \textstyle C=E[(x-f(x))^{2}].
Minimizing this cost produces a value of {\displaystyle \textstyle a} \textstyle a
that is equal to the mean of the data. The cost function can be much more
complicated. Its form depends on the application: for example, in compression it
could be related to the mutual information between {\displaystyle \textstyle x}
\textstyle x and {\displaystyle \textstyle f(x)} \textstyle f(x), whereas in
statistical modeling, it could be related to the posterior probability of the model
given the data (note that in both of those examples those quantities would be
maximized rather than minimized).

Tasks that fall within the paradigm of unsupervised learning are in general
estimation problems; the applications include clustering, the estimation of
statistical distributions, compression and filtering.

Reinforcement learning
See also: Stochastic control
In reinforcement learning, data {\displaystyle \textstyle x} \textstyle x are
usually not given, but generated by an agent's interactions with the environment.
At each point in time {\displaystyle \textstyle t} \textstyle t, the agent performs
an action {\displaystyle \textstyle y_{t}} \textstyle y_{t} and the environment
generates an observation {\displaystyle \textstyle x_{t}} \textstyle x_{t} and an
instantaneous cost {\displaystyle \textstyle c_{t}} \textstyle c_{t}, according to
some (usually unknown) dynamics. The aim is to discover a policy for selecting
actions that minimizes some measure of a long-term cost, e.g., the expected
cumulative cost. The environment's dynamics and the long-term cost for each policy
are usually unknown, but can be estimated.

More formally the environment is modeled as a Markov decision process (MDP) with
states {\displaystyle \textstyle {s_{1},...,s_{n}}\in S} \textstyle
{s_{1},...,s_{n}}\in S and actions {\displaystyle \textstyle {a_{1},...,a_{m}}\in
A} \textstyle {a_{1},...,a_{m}}\in A with the following probability distributions:
the instantaneous cost distribution {\displaystyle \textstyle P(c_{t}|s_{t})}
\textstyle P(c_{t}|s_{t}), the observation distribution {\displaystyle \textstyle
P(x_{t}|s_{t})} \textstyle P(x_{t}|s_{t}) and the transition {\displaystyle
\textstyle P(s_{t+1}|s_{t},a_{t})} \textstyle P(s_{t+1}|s_{t},a_{t}), while a
policy is defined as the conditional distribution over actions given the
observations. Taken together, the two then define a Markov chain (MC). The aim is
to discover the policy (i.e., the MC) that minimizes the cost.

Artificial neural networks are frequently used in reinforcement learning as part of

the overall algorithm.[78][79] Dynamic programming was coupled with Artificial
neural networks (giving neurodynamic programming) by Bertsekas and Tsitsiklis[80]
and applied to multi-dimensional nonlinear problems such as those involved in
vehicle routing,[81] natural resources management[82][83] or medicine[84] because
of the ability of Artificial neural networks to mitigate losses of accuracy even
when reducing the discretization grid density for numerically approximating the
solution of the original control problems.

Tasks that fall within the paradigm of reinforcement learning are control problems,
games and other sequential decision making tasks.

Learning algorithms
See also: Machine learning
Training a neural network model essentially means selecting one model from the set
of allowed models (or, in a Bayesian framework, determining a distribution over the
set of allowed models) that minimizes the cost. Numerous algorithms are available
for training neural network models; most of them can be viewed as a straightforward
application of optimization theory and statistical estimation.

Most employ some form of gradient descent, using backpropagation to compute the
actual gradients. This is done by simply taking the derivative of the cost function
with respect to the network parameters and then changing those parameters in a
gradient-related direction. Backpropagation training algorithms fall into three
categories:

steepest descent (with variable learning rate and momentum, resilient

backpropagation);
quasi-Newton (Broyden-Fletcher-Goldfarb-Shanno, one step secant);
Levenberg-Marquardt and conjugate gradient (Fletcher-Reeves update, Polak-Ribi�re
update, Powell-Beale restart, scaled conjugate gradient).[85]
Evolutionary methods,[86] gene expression programming,[87] simulated annealing,[88]
expectation-maximization, non-parametric methods and particle swarm
optimization[89] are other methods for training neural networks.

Convergent recursive learning algorithm

This is a learning method specially designed for cerebellar model articulation
controller (CMAC) neural networks. In 2004, a recursive least squares algorithm was
introduced to train CMAC neural network online.[90] This algorithm can converge in
one step and update all weights in one step with any new input data. Initially,
this algorithm had computational complexity of O(N3). Based on QR decomposition,
this recursive learning algorithm was simplified to be O(N).[91]

Optimization
The optimization algorithm repeats a two phase cycle, propagation and weight
update. When an input vector is presented to the network, it is propagated forward
through the network, layer by layer, until it reaches the output layer. The output
of the network is then compared to the desired output, using a loss function. The
resulting error value is calculated for each of the neurons in the output layer.
The error values are then propagated from the output back through the network,
until each neuron has an associated error value that reflects its contribution to
the original output.

Backpropagation uses these error values to calculate the gradient of the loss
function. In the second phase, this gradient is fed to the optimization method,
which in turn uses it to update the weights, in an attempt to minimize the loss
function.

Algorithm
Let {\displaystyle N} N be a neural network with {\displaystyle e} e connections,
{\displaystyle m} m inputs, and {\displaystyle n} n outputs.

Below, {\displaystyle x_{1},x_{2},\dots } {\displaystyle x_{1},x_{2},\dots } will

denote vectors in {\displaystyle \mathbb {R} ^{m}} \mathbb {R} ^{m}, {\displaystyle
y_{1},y_{2},\dots } {\displaystyle y_{1},y_{2},\dots } vectors in {\displaystyle
\mathbb {R} ^{n}} \mathbb {R} ^{n}, and {\displaystyle w_{0},w_{1},w_{2},\ldots }
{\displaystyle w_{0},w_{1},w_{2},\ldots } vectors in {\displaystyle \mathbb {R}
^{e}} {\displaystyle \mathbb {R} ^{e}}. These are called inputs, outputs and
weights respectively.

The neural network corresponds to a function {\displaystyle y=f_{N}(w,x)}

{\displaystyle y=f_{N}(w,x)} which, given a weight {\displaystyle w} w, maps an
input {\displaystyle x} x to an output {\displaystyle y} y.

The optimization takes as input a sequence of training examples {\displaystyle

(x_{1},y_{1}),\dots ,(x_{p},y_{p})} {\displaystyle (x_{1},y_{1}),\dots ,
(x_{p},y_{p})} and produces a sequence of weights {\displaystyle
w_{0},w_{1},\dots ,w_{p}} {\displaystyle w_{0},w_{1},\dots ,w_{p}} starting from
some initial weight {\displaystyle w_{0}} w_{0}, usually chosen at random.

These weights are computed in turn: first compute {\displaystyle w_{i}} w_{i} using
only {\displaystyle (x_{i},y_{i},w_{i-1})} {\displaystyle (x_{i},y_{i},w_{i-1})}
for {\displaystyle i=1,\dots ,p} {\displaystyle i=1,\dots ,p}. The output of the
algorithm is then {\displaystyle w_{p}} w_p, giving us a new function
{\displaystyle x\mapsto f_{N}(w_{p},x)} {\displaystyle x\mapsto f_{N}(w_{p},x)}.
The computation is the same in each step, hence only the case {\displaystyle i=1}
i=1 is described.

Calculating {\displaystyle w_{1}} w_{1} from {\displaystyle (x_{1},y_{1},w_{0})}

{\displaystyle (x_{1},y_{1},w_{0})} is done by considering a variable weight
{\displaystyle w} w and applying gradient descent to the function {\displaystyle
w\mapsto E(f_{N}(w,x_{1}),y_{1})} {\displaystyle w\mapsto E(f_{N}(w,x_{1}),y_{1})}
to find a local minimum, starting at {\displaystyle w=w_{0}} {\displaystyle
w=w_{0}}.

This makes {\displaystyle w_{1}} w_{1} the minimizing weight found by gradient
descent.

Algorithm in code

This article's tone or style may not reflect the encyclopedic tone used on
Wikipedia. See Wikipedia's guide to writing better articles for suggestions.
(December 2016) (Learn how and when to remove this template message)
To implement the algorithm above, explicit formulas are required for the gradient
of the function {\displaystyle w\mapsto E(f_{N}(w,x),y)} {\displaystyle w\mapsto
E(f_{N}(w,x),y)} where the function is {\displaystyle E(y,y')=|y-y'|^{2}}
{\displaystyle E(y,y')=|y-y'|^{2}}.

The learning algorithm can be divided into two phases: propagation and weight
update.

Phase 1: propagation
Each propagation involves the following steps:
Propagation forward through the network to generate the output value(s)
Calculation of the cost (error term)
Propagation of the output activations back through the network using the training
pattern target to generate the deltas (the difference between the targeted and
actual output values) of all output and hidden neurons.
Phase 2: weight update
For each weight, the following steps must be followed:

The weight's output delta and input activation are multiplied to find the gradient
of the weight.
A ratio (percentage) of the weight's gradient is subtracted from the weight.
This ratio (percentage) influences the speed and quality of learning; it is called
the learning rate. The greater the ratio, the faster the neuron trains, but the
lower the ratio, the more accurate the training is. The sign of the gradient of a
weight indicates whether the error varies directly with, or inversely to, the
weight. Therefore, the weight must be updated in the opposite direction,
"descending" the gradient.

Learning is repeated (on new batches) until the network performs adequately.

Pseudocode
The following is pseudocode for a stochastic gradient descent algorithm for
training a three-layer network (only one hidden layer):

initialize network weights (often small random values)

do
forEach training example named ex
prediction = neural-net-output(network, ex) // forward pass
actual = teacher-output(ex)
compute error (prediction - actual) at the output units
compute {\displaystyle \Delta w_{h}} \Delta w_h for all weights from hidden
layer to output layer // backward pass
compute {\displaystyle \Delta w_{i}} \Delta w_i for all weights from input
layer to hidden layer // backward pass continued
update network weights // input layer not modified by error estimate
until all examples classified correctly or another stopping criterion satisfied
return the network
The lines labeled "backward pass" can be implemented using the backpropagation
algorithm, which calculates the gradient of the error of the network regarding the
network's modifiable weights.[92]

Extension
The choice of learning rate {\textstyle \eta } {\textstyle \eta } is important,
since a high value can cause too strong a change, causing the minimum to be missed,
while a too low learning rate slows the training unnecessarily.

Optimizations such as Quickprop are primarily aimed at speeding up error

minimization; other improvements mainly try to increase reliability.

Adaptive learning rate

In order to avoid oscillation inside the network such as alternating connection
weights, and to improve the rate of convergence, refinements of this algorithm use
an adaptive learning rate.[93]

Inertia
By using a variable inertia term (Momentum) {\textstyle \alpha } {\textstyle \alpha
} the gradient and the last change can be weighted such that the weight adjustment
additionally depends on the previous change. If the Momentum {\textstyle \alpha }
{\textstyle \alpha } is equal to 0, the change depends solely on the gradient,
while a value of 1 will only depend on the last change.

Similar to a ball rolling down a mountain, whose current speed is determined not
only by the current slope of the mountain but also by its own inertia, inertia can
be added:

{\displaystyle \Delta w_{ij}(t+1)=(1-\alpha )\eta \delta _{j}o_{i}+\alpha \,\Delta

w_{ij}(t)} {\displaystyle \Delta w_{ij}(t+1)=(1-\alpha )\eta \delta _{j}o_{i}
+\alpha \,\Delta w_{ij}(t)}
where:
{\textstyle \Delta w_{ij}(t+1)} {\textstyle \Delta w_{ij}(t+1)} is the change in
weight {\textstyle w_{ij}(t+1)} {\textstyle w_{ij}(t+1)} in the connection of
neuron {\textstyle i} {\textstyle i} to neuron {\textstyle j} {\textstyle j} at
time {\textstyle (t+1),} {\textstyle (t+1),}
{\textstyle \eta } {\textstyle \eta } a learning rate ( {\textstyle \eta <0),}
{\textstyle \eta <0),}
{\textstyle \delta _{j}} {\textstyle \delta _{j}} the error signal of neuron
{\textstyle j} {\textstyle j} and
{\textstyle o_{i}} {\textstyle o_{i}} the output of neuron {\textstyle i}
{\textstyle i}, which is also an input of the current neuron (neuron {\textstyle j}
{\textstyle j}),
{\textstyle \alpha } {\textstyle \alpha } the influence of the inertial term
{\textstyle \Delta w_{ij}(t)} {\textstyle \Delta w_{ij}(t)} (in {\textstyle [0,1]}
{\textstyle [0,1]}). This corresponds to the weight change at the previous point in
time.
Inertia makes the current weight change {\textstyle (t+1)} {\textstyle (t+1)}
depend both on the current gradient of the error function (slope of the mountain,
1st summand), as well as on the weight change from the previous point in time
(inertia, 2nd summand).

With inertia, the problems of getting stuck (in steep ravines and flat plateaus)
are avoided. Since, for example, the gradient of the error function becomes very
small in flat plateaus, a plateau would immediately lead to a "deceleration" of the
gradient descent. This deceleration is delayed by the addition of the inertia term
so that a flat plateau can be escaped more quickly.

Modes of learning
Two modes of learning are available: stochastic and batch. In stochastic learning,
each input creates a weight adjustment. In batch learning weights are adjusted
based on a batch of inputs, accumulating errors over the batch. Stochastic learning
introduces "noise" into the gradient descent process, using the local gradient
calculated from one data point; this reduces the chance of the network getting
stuck in local minima. However, batch learning typically yields a faster, more
stable descent to a local minimum, since each update is performed in the direction
of the average error of the batch. A common compromise choice is to use "mini-
batches", meaning small batches and with samples in each batch selected
stochastically from the entire data set.

Variants
Group method of data handling
Main article: Group method of data handling
The Group Method of Data Handling (GMDH)[94] features fully automatic structural
and parametric model optimization. The node activation functions are Kolmogorov-
Gabor polynomials that permit additions and multiplications. It used a deep
feedforward multilayer perceptron with eight layers.[95] It is a supervised
learning network that grows layer by layer, where each layer is trained by
regression analysis. Useless items are detected using a validation set, and pruned
through regularization. The size and depth of the resulting network depends on the
task.[96]
Convolutional neural networks
Main article: Convolutional neural network
A convolutional neural network (CNN) is a class of deep, feed-forward networks,
composed of one or more convolutional layers with fully connected layers (matching
those in typical Artificial neural networks) on top. It uses tied weights and
pooling layers. In particular, max-pooling[17] is often structured via Fukushima's
convolutional architecture.[97] This architecture allows CNNs to take advantage of
the 2D structure of input data.

CNNs are suitable for processing visual and other two-dimensional data.[98][99]
They have shown superior results in both image and speech applications. They can be
trained with standard backpropagation. CNNs are easier to train than other regular,
deep, feed-forward neural networks and have many fewer parameters to estimate.[100]
Examples of applications in computer vision include DeepDream[101] and robot
navigation.[102]

A recent development has been that of Capsule Neural Network (CapsNet), the idea
behind which is to add structures called capsules to a CNN and to reuse output from
several of those capsules to form more stable (with respect to various
perturbations) representations for higher order capsules.[103]

Long short-term memory

Main article: Long short-term memory
Long short-term memory (LSTM) networks are RNNs that avoid the vanishing gradient
problem.[104] LSTM is normally augmented by recurrent gates called forget gates.
[105] LSTM networks prevent backpropagated errors from vanishing or exploding.[20]
Instead errors can flow backwards through unlimited numbers of virtual layers in
space-unfolded LSTM. That is, LSTM can learn "very deep learning" tasks[9] that
require memories of events that happened thousands or even millions of discrete
time steps ago. Problem-specific LSTM-like topologies can be evolved.[106] LSTM can
handle long delays and signals that have a mix of low and high frequency
components.

Stacks of LSTM RNNs[107] trained by Connectionist Temporal Classification (CTC)

[108] can find an RNN weight matrix that maximizes the probability of the label
sequences in a training set, given the corresponding input sequences. CTC achieves
both alignment and recognition.

In 2003, LSTM started to become competitive with traditional speech recognizers.

[109] In 2007, the combination with CTC achieved first good results on speech data.
[110] In 2009, a CTC-trained LSTM was the first RNN to win pattern recognition
contests, when it won several competitions in connected handwriting recognition.[9]
[35] In 2014, Baidu used CTC-trained RNNs to break the Switchboard Hub5'00 speech
recognition benchmark, without traditional speech processing methods.[111] LSTM
also improved large-vocabulary speech recognition,[112][113] text-to-speech
synthesis,[114] for Google Android,[54][115] and photo-real talking heads.[116] In
2015, Google's speech recognition experienced a 49% improvement through CTC-trained
LSTM.[117]

LSTM became popular in Natural Language Processing. Unlike previous models based on
HMMs and similar concepts, LSTM can learn to recognise context-sensitive languages.
[118] LSTM improved machine translation,[119][120] language modeling[121] and
multilingual language processing.[122] LSTM combined with CNNs improved automatic
image captioning.[123]

Deep reservoir computing

Main article: Reservoir computing
Deep Reservoir Computing and Deep Echo State Networks (deepESNs)[124][125] provide
a framework for efficiently trained models for hierarchical processing of temporal
data, while enabling the investigation of the inherent role of RNN layered
composition.[clarification needed]

Deep belief networks

Main article: Deep belief network

A restricted Boltzmann machine (RBM) with fully connected visible and hidden units.
Note there are no hidden-hidden or visible-visible connections.
A deep belief network (DBN) is a probabilistic, generative model made up of
multiple layers of hidden units. It can be considered a composition of simple
learning modules that make up each layer.[126]

A DBN can be used to generatively pre-train a DNN by using the learned DBN weights
as the initial DNN weights. Backpropagation or other discriminative algorithms can
then tune these weights. This is particularly helpful when training data are
limited, because poorly initialized weights can significantly hinder model
performance. These pre-trained weights are in a region of the weight space that is
closer to the optimal weights than were they randomly chosen. This allows for both
improved modeling and faster convergence of the fine-tuning phase.[127]

Large memory storage and retrieval neural networks

Large memory storage and retrieval neural networks (LAMSTAR)[128][129] are fast
deep learning neural networks of many layers that can use many filters
simultaneously. These filters may be nonlinear, stochastic, logic, non-stationary,
or even non-analytical. They are biologically motivated and learn continuously.

A LAMSTAR neural network may serve as a dynamic neural network in spatial or time
domains or both. Its speed is provided by Hebbian link-weights[130] that integrate
the various and usually different filters (preprocessing functions) into its many
layers and to dynamically rank the significance of the various layers and functions
relative to a given learning task. This grossly imitates biological learning which
integrates various preprocessors (cochlea, retina, etc.) and cortexes (auditory,
visual, etc.) and their various regions. Its deep learning capability is further
enhanced by using inhibition, correlation and its ability to cope with incomplete
data, or "lost" neurons or layers even amidst a task. It is fully transparent due
to its link weights. The link-weights allow dynamic determination of innovation and
redundancy, and facilitate the ranking of layers, of filters or of individual
neurons relative to a task.

LAMSTAR has been applied to many domains, including medical[131][132][133] and

financial predictions,[134] adaptive filtering of noisy speech in unknown noise,
[135] still-image recognition,[136] video image recognition,[137] software
security[138] and adaptive control of non-linear systems.[139] LAMSTAR had a much
faster learning speed and somewhat lower error rate than a CNN based on ReLU-
function filters and max pooling, in 20 comparative studies.[140]

These applications demonstrate delving into aspects of the data that are hidden
from shallow learning networks and the human senses, such as in the cases of
predicting onset of sleep apnea events,[132] of an electrocardiogram of a fetus as
recorded from skin-surface electrodes placed on the mother's abdomen early in
pregnancy,[133] of financial prediction[128] or in blind filtering of noisy speech.
[135]

LAMSTAR was proposed in 1996 (A U.S. Patent 5,920,852 A) and was further developed
Graupe and Kordylewski from 1997�2002.[141][142][143] A modified version, known as
LAMSTAR 2, was developed by Schneider and Graupe in 2008.[144][145]

Stacked (de-noising) auto-encoders

The auto encoder idea is motivated by the concept of a good representation. For
example, for a classifier, a good representation can be defined as one that yields
a better-performing classifier.

An encoder is a deterministic mapping {\displaystyle f_{\theta }} f_{\theta } that

transforms an input vector x into hidden representation y, where {\displaystyle
\theta =\{{\boldsymbol {W}},b\}} \theta =\{{\boldsymbol {W}},b\}, {\displaystyle
{\boldsymbol {W}}} {\boldsymbol {W}} is the weight matrix and b is an offset vector
(bias). A decoder maps back the hidden representation y to the reconstructed input
z via {\displaystyle g_{\theta }} g_{\theta }. The whole process of auto encoding
is to compare this reconstructed input to the original and try to minimize the
error to make the reconstructed value as close as possible to the original.

In stacked denoising auto encoders, the partially corrupted output is cleaned (de-
noised). This idea was introduced in 2010 by Vincent et al.[146] with a specific
approach to good representation, a good representation is one that can be obtained
robustly from a corrupted input and that will be useful for recovering the
corresponding clean input. Implicit in this definition are the following ideas:

The higher level representations are relatively stable and robust to input
corruption;
It is necessary to extract features that are useful for representation of the input
distribution.
The algorithm starts by a stochastic mapping of {\displaystyle {\boldsymbol {x}}}
{\boldsymbol {x}} to {\displaystyle {\tilde {\boldsymbol {x}}}} {\tilde
{\boldsymbol {x}}} through {\displaystyle q_{D}({\tilde {\boldsymbol {x}}}|
{\boldsymbol {x}})} q_{D}({\tilde {\boldsymbol {x}}}|{\boldsymbol {x}}), this is
the corrupting step. Then the corrupted input {\displaystyle {\tilde {\boldsymbol
{x}}}} {\tilde {\boldsymbol {x}}} passes through a basic auto-encoder process and
is mapped to a hidden representation {\displaystyle {\boldsymbol {y}}=f_{\theta }
({\tilde {\boldsymbol {x}}})=s({\boldsymbol {W}}{\tilde {\boldsymbol {x}}}+b)}
{\boldsymbol {y}}=f_{\theta }({\tilde {\boldsymbol {x}}})=s({\boldsymbol {W}}
{\tilde {\boldsymbol {x}}}+b). From this hidden representation, we can reconstruct
{\displaystyle {\boldsymbol {z}}=g_{\theta }({\boldsymbol {y}})} {\boldsymbol
{z}}=g_{\theta }({\boldsymbol {y}}). In the last stage, a minimization algorithm
runs in order to have z as close as possible to uncorrupted input {\displaystyle
{\boldsymbol {x}}} {\boldsymbol {x}}. The reconstruction error {\displaystyle L_{H}
({\boldsymbol {x}},{\boldsymbol {z}})} L_{H}({\boldsymbol {x}},{\boldsymbol {z}})
might be either the cross-entropy loss with an affine-sigmoid decoder, or the
squared error loss with an affine decoder.[146]

In order to make a deep architecture, auto encoders stack.[147] Once the encoding
function {\displaystyle f_{\theta }} f_{\theta } of the first denoising auto
encoder is learned and used to uncorrupt the input (corrupted input), the second
level can be trained.[146]

Once the stacked auto encoder is trained, its output can be used as the input to a
supervised learning algorithm such as support vector machine classifier or a multi-
class logistic regression.[146]

Deep stacking networks

A deep stacking network (DSN)[148] (deep convex network) is based on a hierarchy of
blocks of simplified neural network modules. It was introduced in 2011 by Deng and
Dong.[149] It formulates the learning as a convex optimization problem with a
closed-form solution, emphasizing the mechanism's similarity to stacked
generalization.[150] Each DSN block is a simple module that is easy to train by
itself in a supervised fashion without backpropagation for the entire blocks.[151]

Each block consists of a simplified multi-layer perceptron (MLP) with a single

hidden layer. The hidden layer h has logistic sigmoidal units, and the output layer
has linear units. Connections between these layers are represented by weight matrix
U; input-to-hidden-layer connections have weight matrix W. Target vectors t form
the columns of matrix T, and the input data vectors x form the columns of matrix X.
The matrix of hidden units is {\displaystyle {\boldsymbol {H}}=\sigma ({\boldsymbol
{W}}^{T}{\boldsymbol {X}})} {\boldsymbol {H}}=\sigma ({\boldsymbol {W}}^{T}
{\boldsymbol {X}}). Modules are trained in order, so lower-layer weights W are
known at each stage. The function performs the element-wise logistic sigmoid
operation. Each block estimates the same final label class y, and its estimate is
concatenated with original input X to form the expanded input for the next block.
Thus, the input to the first block contains the original data only, while
downstream blocks' input adds the output of preceding blocks. Then learning the
upper-layer weight matrix U given other weights in the network can be formulated as
a convex optimization problem:

{\displaystyle \min _{U^{T}}f=||{\boldsymbol {U}}^{T}{\boldsymbol {H}}-{\boldsymbol

{T}}||_{F}^{2},} \min _{U^{T}}f=||{\boldsymbol {U}}^{T}{\boldsymbol {H}}-
{\boldsymbol {T}}||_{F}^{2},
which has a closed-form solution.

Unlike other deep architectures, such as DBNs, the goal is not to discover the
transformed feature representation. The structure of the hierarchy of this kind of
architecture makes parallel learning straightforward, as a batch-mode optimization
problem. In purely discriminative tasks, DSNs perform better than conventional
DBNs.[148]

Tensor deep stacking networks

This architecture is a DSN extension. It offers two important improvements: it uses
higher-order information from covariance statistics, and it transforms the non-
convex problem of a lower-layer to a convex sub-problem of an upper-layer.[152]
TDSNs use covariance statistics in a bilinear mapping from each of two distinct
sets of hidden units in the same layer to predictions, via a third-order tensor.

While parallelization and scalability are not considered seriously in conventional

DNNs,[153][154][155] all learning for DSNs and TDSNs is done in batch mode, to
allow parallelization.[149][148] Parallelization allows scaling the design to
larger (deeper) architectures and data sets.

The basic architecture is suitable for diverse tasks such as classification and
regression.

Spike-and-slab RBMs
The need for deep learning with real-valued inputs, as in Gaussian restricted
Boltzmann machines, led to the spike-and-slab RBM (ssRBM), which models continuous-
valued inputs with strictly binary latent variables.[156] Similar to basic RBMs and
its variants, a spike-and-slab RBM is a bipartite graph, while like GRBMs, the
visible units (input) are real-valued. The difference is in the hidden layer, where
each hidden unit has a binary spike variable and a real-valued slab variable. A
spike is a discrete probability mass at zero, while a slab is a density over
continuous domain;[157] their mixture forms a prior.[158]

An extension of ssRBM called �-ssRBM provides extra modeling capacity using

additional terms in the energy function. One of these terms enables the model to
form a conditional distribution of the spike variables by marginalizing out the
slab variables given an observation.

Compound hierarchical-deep models

Compound hierarchical-deep models compose deep networks with non-parametric
Bayesian models. Features can be learned using deep architectures such as DBNs,[25]
DBMs,[159] deep auto encoders,[160] convolutional variants,[161][162] ssRBMs,[157]
deep coding networks,[163] DBNs with sparse feature learning,[164] RNNs,[165]
conditional DBNs,[166] de-noising auto encoders.[167] This provides a better
representation, allowing faster learning and more accurate classification with
high-dimensional data. However, these architectures are poor at learning novel
classes with few examples, because all network units are involved in representing
the input (a distributed representation) and must be adjusted together (high degree
of freedom). Limiting the degree of freedom reduces the number of parameters to
learn, facilitating learning of new classes from few examples. Hierarchical
Bayesian (HB) models allow learning from few examples, for example[168][169][170]
[171][172] for computer vision, statistics and cognitive science.

Compound HD architectures aim to integrate characteristics of both HB and deep

networks. The compound HDP-DBM architecture is a hierarchical Dirichlet process
(HDP) as a hierarchical model, incorporated with DBM architecture. It is a full
generative model, generalized from abstract concepts flowing through the layers of
the model, which is able to synthesize new examples in novel classes that look
"reasonably" natural. All the levels are learned jointly by maximizing a joint log-
probability score.[173]

In a DBM with three hidden layers, the probability of a visible input ? is:

{\displaystyle p({\boldsymbol {\nu }},\psi )={\frac {1}{Z}}\sum _{h}e^{\sum

_{ij}W_{ij}^{(1)}\nu _{i}h_{j}^{1}+\sum _{jl}W_{jl}^{(2)}h_{j}^{1}h_{l}^{2}+\sum
_{lm}W_{lm}^{(3)}h_{l}^{2}h_{m}^{3}},} p({\boldsymbol {\nu }},\psi )={\frac {1}
{Z}}\sum _{h}e^{\sum _{ij}W_{ij}^{(1)}\nu _{i}h_{j}^{1}+\sum
_{jl}W_{jl}^{(2)}h_{j}^{1}h_{l}^{2}+\sum _{lm}W_{lm}^{(3)}h_{l}^{2}h_{m}^{3}},
where {\displaystyle {\boldsymbol {h}}=\{{\boldsymbol {h}}^{(1)},{\boldsymbol
{h}}^{(2)},{\boldsymbol {h}}^{(3)}\}} {\boldsymbol {h}}=\{{\boldsymbol {h}}^{(1)},
{\boldsymbol {h}}^{(2)},{\boldsymbol {h}}^{(3)}\} is the set of hidden units, and
{\displaystyle \psi =\{{\boldsymbol {W}}^{(1)},{\boldsymbol {W}}^{(2)},{\boldsymbol
{W}}^{(3)}\}} \psi =\{{\boldsymbol {W}}^{(1)},{\boldsymbol {W}}^{(2)},{\boldsymbol
{W}}^{(3)}\} are the model parameters, representing visible-hidden and hidden-
hidden symmetric interaction terms.

A learned DBM model is an undirected model that defines the joint distribution
{\displaystyle P(\nu ,h^{1},h^{2},h^{3})} P(\nu ,h^{1},h^{2},h^{3}). One way to
express what has been learned is the conditional model {\displaystyle P(\nu
,h^{1},h^{2}|h^{3})} P(\nu ,h^{1},h^{2}|h^{3}) and a prior term {\displaystyle
P(h^{3})} P(h^{3}).

Here {\displaystyle P(\nu ,h^{1},h^{2}|h^{3})} P(\nu ,h^{1},h^{2}|h^{3}) represents

a conditional DBM model, which can be viewed as a two-layer DBM but with bias terms
given by the states of {\displaystyle h^{3}} h^{3}:

{\displaystyle P(\nu ,h^{1},h^{2}|h^{3})={\frac {1}{Z(\psi ,h^{3})}}e^{\sum

_{ij}W_{ij}^{(1)}\nu _{i}h_{j}^{1}+\sum _{jl}W_{jl}^{(2)}h_{j}^{1}h_{l}^{2}+\sum
_{lm}W_{lm}^{(3)}h_{l}^{2}h_{m}^{3}}.} P(\nu ,h^{1},h^{2}|h^{3})={\frac {1}
{Z(\psi ,h^{3})}}e^{\sum _{ij}W_{ij}^{(1)}\nu _{i}h_{j}^{1}+\sum
_{jl}W_{jl}^{(2)}h_{j}^{1}h_{l}^{2}+\sum _{lm}W_{lm}^{(3)}h_{l}^{2}h_{m}^{3}}.
Deep predictive coding networks
A deep predictive coding network (DPCN) is a predictive coding scheme that uses
top-down information to empirically adjust the priors needed for a bottom-up
inference procedure by means of a deep, locally connected, generative model. This
works by extracting sparse features from time-varying observations using a linear
dynamical model. Then, a pooling strategy is used to learn invariant feature
representations. These units compose to form a deep architecture and are trained by
greedy layer-wise unsupervised learning. The layers constitute a kind of Markov
chain such that the states at any layer depend only on the preceding and succeeding
layers.

DPCNs predict the representation of the layer, by using a top-down approach using
the information in upper layer and temporal dependencies from previous states.[174]

DPCNs can be extended to form a convolutional network.[174]

Networks with separate memory structures

Integrating external memory with Artificial neural networks dates to early research
in distributed representations[175] and Kohonen's self-organizing maps. For
example, in sparse distributed memory or hierarchical temporal memory, the patterns
encoded by neural networks are used as addresses for content-addressable memory,
with "neurons" essentially serving as address encoders and decoders. However, the
early controllers of such memories were not differentiable.

LSTM-related differentiable memory structures

Apart from long short-term memory (LSTM), other approaches also added
differentiable memory to recurrent functions. For example:

Differentiable push and pop actions for alternative memory networks called neural
stack machines[176][177]
Memory networks where the control network's external differentiable storage is in
the fast weights of another network[178]
LSTM forget gates[179]
Self-referential RNNs with special output units for addressing and rapidly
manipulating the RNN's own weights in differentiable fashion (internal storage)
[180][181]
Learning to transduce with unbounded memory[182]
Neural Turing machines
Main article: Neural Turing machine
Neural Turing machines[183] couple LSTM networks to external memory resources, with
which they can interact by attentional processes. The combined system is analogous
to a Turing machine but is differentiable end-to-end, allowing it to be efficiently
trained by gradient descent. Preliminary results demonstrate that neural Turing
machines can infer simple algorithms such as copying, sorting and associative
recall from input and output examples.

Differentiable neural computers (DNC) are an NTM extension. They out-performed

Neural turing machines, long short-term memory systems and memory networks on
sequence-processing tasks.[184][185][186][187][188]

Semantic hashing
Approaches that represent previous experiences directly and use a similar
experience to form a local model are often called nearest neighbour or k-nearest
neighbors methods.[189] Deep learning is useful in semantic hashing[190] where a
deep graphical model the word-count vectors[191] obtained from a large set of
documents.[clarification needed] Documents are mapped to memory addresses in such a
way that semantically similar documents are located at nearby addresses. Documents
similar to a query document can then be found by accessing all the addresses that
differ by only a few bits from the address of the query document. Unlike sparse
distributed memory that operates on 1000-bit addresses, semantic hashing works on
32 or 64-bit addresses found in a conventional computer architecture.

Memory networks
Memory networks[192][193] are another extension to neural networks incorporating
long-term memory. The long-term memory can be read and written to, with the goal of
using it for prediction. These models have been applied in the context of question
answering (QA) where the long-term memory effectively acts as a (dynamic) knowledge
base and the output is a textual response.[194] A team of electrical and computer
engineers from UCLA Samueli School of Engineering has created a physical artificial
neural network that can analyze large volumes of data and identify objects at the
actual speed of light.[195]

Pointer networks
Deep neural networks can be potentially improved by deepening and parameter
reduction, while maintaining trainability. While training extremely deep (e.g., 1
million layers) neural networks might not be practical, CPU-like architectures such
as pointer networks[196] and neural random-access machines[197] overcome this
limitation by using external random-access memory and other components that
typically belong to a computer architecture such as registers, ALU and pointers.
Such systems operate on probability distribution vectors stored in memory cells and
registers. Thus, the model is fully differentiable and trains end-to-end. The key
characteristic of these models is that their depth, the size of their short-term
memory, and the number of parameters can be altered independently � unlike models
like LSTM, whose number of parameters grows quadratically with memory size.

Encoder�decoder networks
Encoder�decoder frameworks are based on neural networks that map highly structured
input to highly structured output. The approach arose in the context of machine
translation,[198][199][200] where the input and output are written sentences in two
natural languages. In that work, an LSTM RNN or CNN was used as an encoder to
summarize a source sentence, and the summary was decoded using a conditional RNN
language model to produce the translation.[201] These systems share building
blocks: gated RNNs and CNNs and trained attention mechanisms.

Multilayer kernel machine

Multilayer kernel machines (MKM) are a way of learning highly nonlinear functions
by iterative application of weakly nonlinear kernels. They use the kernel principal
component analysis (KPCA),[202] as a method for the unsupervised greedy layer-wise
pre-training step of deep learning.[203]

Layer {\displaystyle l+1} l+1 learns the representation of the previous layer
{\displaystyle l} l, extracting the {\displaystyle n_{l}} n_{l} principal component
(PC) of the projection layer {\displaystyle l} l output in the feature domain
induced by the kernel. For the sake of dimensionality reduction of the updated
representation in each layer, a supervised strategy selects the best informative
features among features extracted by KPCA. The process is:

rank the {\displaystyle n_{l}} n_{l} features according to their mutual information
with the class labels;
for different values of K and {\displaystyle m_{l}\in \{1,\ldots ,n_{l}\}} m_{l}\in
\{1,\ldots ,n_{l}\}, compute the classification error rate of a K-nearest neighbor
(K-NN) classifier using only the {\displaystyle m_{l}} m_{l} most informative
features on a validation set;
the value of {\displaystyle m_{l}} m_{l} with which the classifier has reached the
lowest error rate determines the number of features to retain.
Some drawbacks accompany the KPCA method as the building cells of an MKM.

A more straightforward way to use kernel machines for deep learning was developed
for spoken language understanding.[204] The main idea is to use a kernel machine to
approximate a shallow neural net with an infinite number of hidden units, then use
stacking to splice the output of the kernel machine and the raw input in building
the next, higher level of the kernel machine. The number of levels in the deep
convex network is a hyper-parameter of the overall system, to be determined by
cross validation.

Neural architecture search

Main article: Neural architecture search
Neural architecture search (NAS) uses machine learning to automate the design of
Artificial neural networks. Various approaches to NAS have designed networks that
compare well with hand-designed systems. The basic search algorithm is to propose a
candidate model, evaluate it against a dataset and use the results as feedback to
teach the NAS network.[205]

Use
Using Artificial neural networks requires an understanding of their
characteristics.

Choice of model: This depends on the data representation and the application.
Overly complex models slow learning.
Learning algorithm: Numerous trade-offs exist between learning algorithms. Almost
any algorithm will work well with the correct hyperparameters for training on a
particular data set. However, selecting and tuning an algorithm for training on
unseen data requires significant experimentation.
Robustness: If the model, cost function and learning algorithm are selected
appropriately, the resulting ANN can become robust.
ANN capabilities fall within the following broad categories:[citation needed]

Function approximation, or regression analysis, including time series prediction,

fitness approximation and modeling.
Classification, including pattern and sequence recognition, novelty detection and
sequential decision making.
Data processing, including filtering, clustering, blind source separation and
compression.
Robotics, including directing manipulators and prostheses.
Control, including computer numerical control.
Applications
Because of their ability to reproduce and model nonlinear processes, Artificial
neural networks have found many applications in a wide range of disciplines.

Application areas include system identification and control (vehicle control,

trajectory prediction,[206] process control, natural resource management), quantum
chemistry,[207] general game playing,[208] pattern recognition (radar systems, face
identification, signal classification,[209] 3D reconstruction,[210] object
recognition and more), sequence recognition (gesture, speech, handwritten and
printed text recognition), medical diagnosis, finance[211] (e.g. automated trading
systems), data mining, visualization, machine translation, social network
filtering[212] and e-mail spam filtering.

Artificial neural networks have been used to diagnose cancers, including lung
cancer,[213] prostate cancer, colorectal cancer[214] and to distinguish highly
invasive cancer cell lines from less invasive lines using only cell shape
information.[215][216]

Artificial neural networks have been used to accelerate reliability analysis of

infrastructures subject to natural disasters[217][218] and to predict foundation
settlements.[219]

Artificial neural networks have also been used for building black-box models in
geoscience: hydrology,[220][221] ocean modelling and coastal engineering,[222][223]
and geomorphology.[224]

Artificial neural networks have been employed with some success also in
cybersecurity, with the objective to discriminate between legitimate activities and
malicious ones. For example, machine learning has been used for classifying android
malware,[225] for identifying domains belonging to threat actors[226] and for
detecting URLs posing a security risk.[227] Research is being carried out also on
ANN systems designed for penetration testing,[228] for detecting botnets,[229]
credit cards frauds,[230] network intrusions and, more in general, potentially
infected machines.

Artificial neural networks have been proposed as a tool to simulate the properties
of many-body open quantum systems.[231][232][233][234]

Types of models
Many types of models are used, defined at different levels of abstraction and
modeling different aspects of neural systems. They range from models of the short-
term behavior of individual neurons,[235] models of how the dynamics of neural
circuitry arise from interactions between individual neurons and finally to models
of how behavior can arise from abstract neural modules that represent complete
subsystems. These include models of the long-term, and short-term plasticity, of
neural systems and their relations to learning and memory from the individual
neuron to the system level.

Theoretical properties
Computational power
The multilayer perceptron is a universal function approximator, as proven by the
universal approximation theorem. However, the proof is not constructive regarding
the number of neurons required, the network topology, the weights and the learning
parameters.

A specific recurrent architecture with rational valued weights (as opposed to full
precision real number-valued weights) has the full power of a universal Turing
machine,[236] using a finite number of neurons and standard linear connections.
Further, the use of irrational values for weights results in a machine with super-
Turing power.[237]

Capacity
Models' "capacity" property roughly corresponds to their ability to model any given
function. It is related to the amount of information that can be stored in the
network and to the notion of complexity.[citation needed]

Convergence
Models may not consistently converge on a single solution, firstly because many
local minima may exist, depending on the cost function and the model. Secondly, the
optimization method used might not guarantee to converge when it begins far from
any local minimum. Thirdly, for sufficiently large data or parameters, some methods
become impractical. However, for CMAC neural network, a recursive least squares
algorithm was introduced to train it, and this algorithm can be guaranteed to
converge in one step.[90]

Generalization and statistics

Applications whose goal is to create a system that generalizes well to unseen
examples, face the possibility of over-training. This arises in convoluted or over-
specified systems when the capacity of the network significantly exceeds the needed
free parameters. Two approaches address over-training. The first is to use cross-
validation and similar techniques to check for the presence of over-training and
optimally select hyperparameters to minimize the generalization error. The second
is to use some form of regularization. This concept emerges in a probabilistic
(Bayesian) framework, where regularization can be performed by selecting a larger
prior probability over simpler models; but also in statistical learning theory,
where the goal is to minimize over two quantities: the 'empirical risk' and the
'structural risk', which roughly corresponds to the error over the training set and
the predicted error in unseen data due to overfitting.
Confidence analysis of a neural network
Supervised neural networks that use a mean squared error (MSE) cost function can
use formal statistical methods to determine the confidence of the trained model.
The MSE on a validation set can be used as an estimate for variance. This value can
then be used to calculate the confidence interval of the output of the network,
assuming a normal distribution. A confidence analysis made this way is
statistically valid as long as the output probability distribution stays the same
and the network is not modified.

By assigning a softmax activation function, a generalization of the logistic

function, on the output layer of the neural network (or a softmax component in a
component-based neural network) for categorical target variables, the outputs can
be interpreted as posterior probabilities. This is very useful in classification as
it gives a certainty measure on classifications.

The softmax activation function is:

{\displaystyle y_{i}={\frac {e^{x_{i}}}{\sum _{j=1}^{c}e^{x_{j}}}}} y_{i}={\frac

{e^{x_{i}}}{\sum _{j=1}^{c}e^{x_{j}}}}

Criticism
Training issues
A common criticism of neural networks, particularly in robotics, is that they
require too much training for real-world operation.[citation needed] Potential
solutions include randomly shuffling training examples, by using a numerical
optimization algorithm that does not take too large steps when changing the network
connections following an example and by grouping examples in so-called mini-
batches. Improving the training efficiency and convergence capability has always
been an ongoing research area for neural network. For example, by introducing a
recursive least squares algorithm for CMAC neural network, the training process
only takes one step to converge.[90]

Theoretical issues
A fundamental objection is that they do not reflect how real neurons function. Back
propagation is a critical part of most artificial neural networks, although no such
mechanism exists in biological neural networks.[238] How information is coded by
real neurons is not known. Sensor neurons fire action potentials more frequently
with sensor activation and muscle cells pull more strongly when their associated
motor neurons receive action potentials more frequently.[239] Other than the case
of relaying information from a sensor neuron to a motor neuron, almost nothing of
the principles of how information is handled by biological neural networks is
known. This is a subject of active research in neural coding.

The motivation behind artificial neural networks is not necessarily to strictly

replicate neural function, but to use biological neural networks as an inspiration.
A central claim of artificial neural networks is therefore that it embodies some
new and powerful general principle for processing information. Unfortunately, these
general principles are ill-defined. It is often claimed that they are emergent from
the network itself. This allows simple statistical association (the basic function
of artificial neural networks) to be described as learning or recognition.
Alexander Dewdney commented that, as a result, artificial neural networks have a
"something-for-nothing quality, one that imparts a peculiar aura of laziness and a
distinct lack of curiosity about just how good these computing systems are. No
human hand (or mind) intervenes; solutions are found as if by magic; and no one, it
seems, has learned anything".[240]

Biological brains use both shallow and deep circuits as reported by brain anatomy,
[241] displaying a wide variety of invariance. Weng[242] argued that the brain
self-wires largely according to signal statistics and therefore, a serial cascade
cannot catch all major statistical dependencies.

Hardware issues
Large and effective neural networks require considerable computing resources.[243]
While the brain has hardware tailored to the task of processing signals through a
graph of neurons, simulating even a simplified neuron on von Neumann architecture
may compel a neural network designer to fill many millions of database rows for its
connections � which can consume vast amounts of memory and storage. Furthermore,
the designer often needs to transmit signals through many of these connections and
their associated neurons � which must often be matched with enormous CPU processing
power and time.

Schmidhuber notes that the resurgence of neural networks in the twenty-first

century is largely attributable to advances in hardware: from 1991 to 2015,
computing power, especially as delivered by GPGPUs (on GPUs), has increased around
a million-fold, making the standard backpropagation algorithm feasible for training
networks that are several layers deeper than before.[244] The use of accelerators
such as FPGAs and GPUs can reduce training times from months to days.[245][243]

Neuromorphic engineering addresses the hardware difficulty directly, by

constructing non-von-Neumann chips to directly implement neural networks in
circuitry. Another chip optimized for neural network processing is called a Tensor
Processing Unit, or TPU.[246]

Practical counterexamples to criticisms

Arguments against Dewdney's position are that neural networks have been
successfully used to solve many complex and diverse tasks, ranging from
autonomously flying aircraft[247] to detecting credit card fraud to mastering the
game of Go.

Technology writer Roger Bridgman commented:

Neural networks, for instance, are in the dock not only because they have been
hyped to high heaven, (what hasn't?) but also because you could create a successful
net without understanding how it worked: the bunch of numbers that captures its
behaviour would in all probability be "an opaque, unreadable table...valueless as a
scientific resource".

In spite of his emphatic declaration that science is not technology, Dewdney seems
here to pillory neural nets as bad science when most of those devising them are
just trying to be good engineers. An unreadable table that a useful machine could
read would still be well worth having.[248]

Although it is true that analyzing what has been learned by an artificial neural
network is difficult, it is much easier to do so than to analyze what has been
learned by a biological neural network. Furthermore, researchers involved in
exploring learning algorithms for neural networks are gradually uncovering general
principles that allow a learning machine to be successful. For example, local vs
non-local learning and shallow vs deep architecture.[249]

Hybrid approaches
Advocates of hybrid models (combining neural networks and symbolic approaches),
claim that such a mixture can better capture the mechanisms of the human mind.[250]
[251]

Types
Main article: Types of artificial neural networks
Artificial neural networks have many variations. The simplest, static types have
one or more static components, including number of units, number of layers, unit
weights and topology. Dynamic types allow one or more of these to change during the
learning process. The latter are much more complicated, but can shorten learning
periods and produce better results. Some types allow/require learning to be
"supervised" by the operator, while others operate independently. Some types
operate purely in hardware, while others are purely software and run on general
purpose computers.

Gallery

A single-layer feedforward artificial neural network. Arrows originating from

{\displaystyle \scriptstyle x_{2}} \scriptstyle x_{2} are omitted for clarity.
There are p inputs to this network and q outputs. In this system, the value of the
qth output, {\displaystyle \scriptstyle y_{q}} \scriptstyle y_{q} would be
calculated as {\displaystyle \scriptstyle y_{q}=K*(\sum (x_{i}*w_{iq})-b_{q})}
{\displaystyle \scriptstyle y_{q}=K*(\sum (x_{i}*w_{iq})-b_{q})}

A two-layer feedforward artificial neural network.

An artificial neural network.

An ANN dependency graph.

A single-layer feedforward artificial neural network with 4 inputs, 6 hidden and 2

outputs. Given position state and direction outputs wheel based control values.

A two-layer feedforward artificial neural network with 8 inputs, 2x8 hidden and 2
outputs. Given position state, direction and other environment values outputs
thruster based control values.

Parallel pipeline structure of CMAC neural network. This learning algorithm can
converge in one step.

See also

This "see also" section may contain an excessive number of suggestions. Please
ensure that only the most relevant links are given, that they are not red links,
and that any links are not already in this article. (March 2018) (Learn how and
when to remove this template message)
Hierarchical temporal memory
20Q
ADALINE
Adaptive resonance theory
Artificial life
Associative memory
Autoencoder
BEAM robotics
Biological cybernetics
Biologically inspired computing
Blue Brain Project
Catastrophic interference
Cerebellar Model Articulation Controller (CMAC)
Cognitive architecture
Cognitive science
Convolutional neural network (CNN)
Connectionist expert system
Connectomics
Cultured neuronal networks
Deep learning
Encog
Fuzzy logic
Gene expression programming
Genetic algorithm
Genetic programming
Group method of data handling
Habituation
In Situ Adaptive Tabulation
Machine learning concepts
Models of neural computation
Neuroevolution
Neural coding
Neural gas
Neural machine translation
Neural network software
Neuroscience
Nonlinear system identification
Optical neural network
Parallel Constraint Satisfaction Processes
Parallel distributed processing
Radial basis function network
Recurrent neural networks
Self-organizing map
Spiking neural network
Systolic array
Tensor product network
Time delay neural network (TDNN)
References
McCulloch, Warren; Walter Pitts (1943). "A Logical Calculus of Ideas Immanent in
Nervous Activity". Bulletin of Mathematical Biophysics. 5 (4): 115�133.
doi:10.1007/BF02478259.
Kleene, S.C. (1956). "Representation of Events in Nerve Nets and Finite Automata".
Annals of Mathematics Studies (34). Princeton University Press. pp. 3�41. Retrieved
17 June 2017.
Hebb, Donald (1949). The Organization of Behavior. New York: Wiley. ISBN 978-1-
135-63190-1.
Farley, B.G.; W.A. Clark (1954). "Simulation of Self-Organizing Systems by Digital
Computer". IRE Transactions on Information Theory. 4 (4): 76�84.
doi:10.1109/TIT.1954.1057468.
Rochester, N.; J.H. Holland; L.H. Habit; W.L. Duda (1956). "Tests on a cell
assembly theory of the action of the brain, using a large digital computer". IRE
Transactions on Information Theory. 2 (3): 80�93. doi:10.1109/TIT.1956.1056810.
Rosenblatt, F. (1958). "The Perceptron: A Probabilistic Model For Information
Storage And Organization In The Brain". Psychological Review. 65 (6): 386�408.
CiteSeerX 10.1.1.588.3775. doi:10.1037/h0042519. PMID 13602029.
Werbos, P.J. (1975). Beyond Regression: New Tools for Prediction and Analysis in
the Behavioral Sciences.
David H. Hubel and Torsten N. Wiesel (2005). Brain and visual perception: the
story of a 25-year collaboration. Oxford University Press US. p. 106. ISBN 978-0-
19-517618-6.
 Schmidhuber, J. (2015). "Deep Learning in Neural Networks: An Overview". Neural
Networks. 61: 85�117. arXiv:1404.7828. doi:10.1016/j.neunet.2014.09.003. PMID
25462637.
Ivakhnenko, A. G. (1973). Cybernetic Predicting Devices. CCM Information
Corporation.
Ivakhnenko, A. G.; Grigor'evich Lapa, Valentin (1967). Cybernetics and forecasting
techniques. American Elsevier Pub. Co.
Minsky, Marvin; Papert, Seymour (1969). Perceptrons: An Introduction to
Computational Geometry. MIT Press. ISBN 978-0-262-63022-1.
Rumelhart, D.E; McClelland, James (1986). Parallel Distributed Processing:
Explorations in the Microstructure of Cognition. Cambridge: MIT Press. ISBN 978-0-
262-63110-5.
Qian, N.; Sejnowski, T.J. (1988). "Predicting the secondary structure of globular
proteins using neural network models" (PDF). Journal of Molecular Biology. 202:
865�884. Qian1988.
Rost, B.; Sander, C. (1993). "Prediction of protein secondary structure at better
than 70% accuracy" (PDF). Journal of Molecular Biology. 232: 584�599. Rost1993.
J. Weng, N. Ahuja and T. S. Huang, "Cresceptron: a self-organizing neural network
which grows adaptively," Proc. International Joint Conference on Neural Networks,
Baltimore, Maryland, vol I, pp. 576�581, June, 1992.
J. Weng, N. Ahuja and T. S. Huang, "Learning recognition and segmentation of 3-D
objects from 2-D images," Proc. 4th International Conf. Computer Vision, Berlin,
Germany, pp. 121�128, May, 1993.
J. Weng, N. Ahuja and T. S. Huang, "Learning recognition and segmentation using
the Cresceptron," International Journal of Computer Vision, vol. 25, no. 2, pp.
105�139, Nov. 1997.
Dominik Scherer, Andreas C. M�ller, and Sven Behnke: "Evaluation of Pooling
Operations in Convolutional Architectures for Object Recognition," In 20th
International Conference Artificial Neural Networks (ICANN), pp. 92�101, 2010.
doi:10.1007/978-3-642-15825-4_10.
S. Hochreiter., "Untersuchungen zu dynamischen neuronalen Netzen," Diploma thesis.
Institut f. Informatik, Technische Univ. Munich. Advisor: J. Schmidhuber, 1991.
Hochreiter, S.; et al. (15 January 2001). "Gradient flow in recurrent nets: the
difficulty of learning long-term dependencies". In Kolen, John F.; Kremer, Stefan
C. (eds.). A Field Guide to Dynamical Recurrent Networks. John Wiley & Sons. ISBN
978-0-7803-5369-5.
J. Schmidhuber., "Learning complex, extended sequences using the principle of
history compression," Neural Computation, 4, pp. 234�242, 1992.
Sven Behnke (2003). Hierarchical Neural Networks for Image Interpretation (PDF).
Lecture Notes in Computer Science. 2766. Springer.
Smolensky, P. (1986). "Information processing in dynamical systems: Foundations of
harmony theory.". In D. E. Rumelhart; J. L. McClelland; PDP Research Group (eds.).
Parallel Distributed Processing: Explorations in the Microstructure of Cognition.
1. pp. 194�281. ISBN 9780262680530.
Hinton, G. E.; Osindero, S.; Teh, Y. (2006). "A fast learning algorithm for deep
belief nets" (PDF). Neural Computation. 18 (7): 1527�1554. CiteSeerX
10.1.1.76.1541. doi:10.1162/neco.2006.18.7.1527. PMID 16764513.
Hinton, G. (2009). "Deep belief networks". Scholarpedia. 4 (5): 5947.
Bibcode:2009SchpJ...4.5947H. doi:10.4249/scholarpedia.5947.
Ng, Andrew; Dean, Jeff (2012). "Building High-level Features Using Large Scale
Unsupervised Learning". arXiv:1112.6209 [cs.LG].
Yang, J. J.; Pickett, M. D.; Li, X. M.; Ohlberg, D. A. A.; Stewart, D. R.;
Williams, R. S. (2008). "Memristive switching mechanism for metal/oxide/metal
nanodevices". Nat. Nanotechnol. 3 (7): 429�433. doi:10.1038/nnano.2008.160. PMID
18654568.
Strukov, D. B.; Snider, G. S.; Stewart, D. R.; Williams, R. S. (2008). "The
missing memristor found". Nature. 453 (7191): 80�83. Bibcode:2008Natur.453...80S.
doi:10.1038/nature06932. PMID 18451858.
Ciresan, Dan Claudiu; Meier, Ueli; Gambardella, Luca Maria; Schmidhuber, J�rgen
(21 September 2010). "Deep, Big, Simple Neural Nets for Handwritten Digit
Recognition". Neural Computation. 22 (12): 3207�3220. arXiv:1003.0358.
doi:10.1162/neco_a_00052. ISSN 0899-7667. PMID 20858131.
2012 Kurzweil AI Interview with J�rgen Schmidhuber on the eight competitions won
by his Deep Learning team 2009�2012
"How bio-inspired deep learning keeps winning competitions | KurzweilAI".
www.kurzweilai.net. Retrieved 16 June 2017.
Graves, Alex; and Schmidhuber, J�rgen; Offline Handwriting Recognition with
Multidimensional Recurrent Neural Networks, in Bengio, Yoshua; Schuurmans, Dale;
Lafferty, John; Williams, Chris K. I.; and Culotta, Aron (eds.), Advances in Neural
Information Processing Systems 22 (NIPS'22), 7�10 December 2009, Vancouver, BC,
Neural Information Processing Systems (NIPS) Foundation, 2009, pp. 545�552.
Graves, A.; Liwicki, M.; Fernandez, S.; Bertolami, R.; Bunke, H.; Schmidhuber, J.
(2009). "A Novel Connectionist System for Improved Unconstrained Handwriting
Recognition" (PDF). IEEE Transactions on Pattern Analysis and Machine Intelligence.
31 (5): 855�868. CiteSeerX 10.1.1.139.4502. doi:10.1109/tpami.2008.137. PMID
19299860.
Graves, Alex; Schmidhuber, J�rgen (2009). Bengio, Yoshua; Schuurmans, Dale;
Lafferty, John; Williams, Chris editor-K. I.; Culotta, Aron (eds.). "Offline
Handwriting Recognition with Multidimensional Recurrent Neural Networks". Neural
Information Processing Systems (NIPS) Foundation. Curran Associates, Inc: 545�552.
Graves, A.; Liwicki, M.; Fern�ndez, S.; Bertolami, R.; Bunke, H.; Schmidhuber, J.
(May 2009). "A Novel Connectionist System for Unconstrained Handwriting
Recognition". IEEE Transactions on Pattern Analysis and Machine Intelligence. 31
(5): 855�868. CiteSeerX 10.1.1.139.4502. doi:10.1109/tpami.2008.137. ISSN 0162-
8828. PMID 19299860.
Ciresan, Dan; Meier, Ueli; Masci, Jonathan; Schmidhuber, J�rgen (August 2012).
"Multi-column deep neural network for traffic sign classification". Neural
Networks. Selected Papers from IJCNN 2011. 32: 333�338. CiteSeerX 10.1.1.226.8219.
doi:10.1016/j.neunet.2012.02.023. PMID 22386783.
Ciresan, Dan; Giusti, Alessandro; Gambardella, Luca M.; Schmidhuber, Juergen
(2012). Pereira, F.; Burges, C. J. C.; Bottou, L.; Weinberger, K. Q. (eds.).
Advances in Neural Information Processing Systems 25 (PDF). Curran Associates, Inc.
pp. 2843�2851.
Ciresan, Dan; Meier, U.; Schmidhuber, J. (June 2012). Multi-column deep neural
networks for image classification. 2012 IEEE Conference on Computer Vision and
Pattern Recognition. pp. 3642�3649. arXiv:1202.2745. Bibcode:2012arXiv1202.2745C.
CiteSeerX 10.1.1.300.3283. doi:10.1109/cvpr.2012.6248110. ISBN 978-1-4673-1228-8.
Ciresan, D. C.; Meier, U.; Masci, J.; Gambardella, L. M.; Schmidhuber, J. (2011).
"Flexible, High Performance Convolutional Neural Networks for Image Classification"
(PDF). International Joint Conference on Artificial Intelligence. doi:10.5591/978-
1-57735-516-8/ijcai11-210.
Krizhevsky, Alex; Sutskever, Ilya; Hinton, Geoffry (2012). "ImageNet
Classification with Deep Convolutional Neural Networks" (PDF). NIPS 2012: Neural
Information Processing Systems, Lake Tahoe, Nevada.
Fukushima, K. (1980). "Neocognitron: A self-organizing neural network model for a
mechanism of pattern recognition unaffected by shift in position". Biological
Cybernetics. 36 (4): 93�202. doi:10.1007/BF00344251. PMID 7370364.
Riesenhuber, M; Poggio, T (1999). "Hierarchical models of object recognition in
cortex". Nature Neuroscience. 2 (11): 1019�1025. doi:10.1038/14819. PMID 10526343.
Hinton, Geoffrey (31 May 2009). "Deep belief networks". Scholarpedia. 4 (5): 5947.
Bibcode:2009SchpJ...4.5947H. doi:10.4249/scholarpedia.5947. ISSN 1941-6016.
Markoff, John (23 November 2012). "Scientists See Promise in Deep-Learning
Programs". New York Times.
Martines, H.; Bengio, Y.; Yannakakis, G. N. (2013). "Learning Deep Physiological
Models of Affect". IEEE Computational Intelligence Magazine (Submitted manuscript).
8 (2): 20�33. doi:10.1109/mci.2013.2247823.
J. Weng, "Why Have We Passed 'Neural Networks Do not Abstract Well'?," Natural
Intelligence: the INNS Magazine, vol. 1, no.1, pp. 13�22, 2011.
 Z. Ji, J. Weng, and D. Prokhorov, "Where-What Network 1: Where and What Assist
Each Other Through Top-down Connections," Proc. 7th International Conference on
Development and Learning (ICDL'08), Monterey, CA, Aug. 9�12, pp. 1�6, 2008.
X. Wu, G. Guo, and J. Weng, "Skull-closed Autonomous Development: WWN-7 Dealing
with Scales," Proc. International Conference on Brain-Mind, July 27�28, East
Lansing, Michigan, pp. 1�9, 2013.
Zell, Andreas (1994). "chapter 5.2". Simulation Neuronaler Netze [Simulation of
Neural Networks] (in German) (1st ed.). Addison-Wesley. ISBN 978-3-89319-554-1.
Abbod, Maysam F (2007). "Application of Artificial Intelligence to the Management
of Urological Cancer". The Journal of Urology. 178 (4): 1150�1156.
doi:10.1016/j.juro.2007.05.122. PMID 17698099.
DAWSON, CHRISTIAN W (1998). "An artificial neural network approach to rainfall-
runoff modelling". Hydrological Sciences Journal. 43 (1): 47�66.
doi:10.1080/02626669809492102.
"The Machine Learning Dictionary".
Schmidhuber, J�rgen (2015). "Deep Learning". Scholarpedia. 10 (11): 32832.
Bibcode:2015SchpJ..1032832S. doi:10.4249/scholarpedia.32832.
Dreyfus, Stuart E. (1 September 1990). "Artificial neural networks, back
propagation, and the Kelley-Bryson gradient procedure". Journal of Guidance,
Control, and Dynamics. 13 (5): 926�928. Bibcode:1990JGCD...13..926D.
doi:10.2514/3.25422. ISSN 0731-5090.
Eiji Mizutani, Stuart Dreyfus, Kenichi Nishio (2000). On derivation of MLP
backpropagation from the Kelley-Bryson optimal-control gradient formula and its
application. Proceedings of the IEEE International Joint Conference on Neural
Networks (IJCNN 2000), Como Italy, July 2000. Online
Kelley, Henry J. (1960). "Gradient theory of optimal flight paths". ARS Journal.
30 (10): 947�954. doi:10.2514/8.5282.
Arthur E. Bryson (1961, April). A gradient method for optimizing multi-stage
allocation processes. In Proceedings of the Harvard Univ. Symposium on digital
computers and their applications.
Dreyfus, Stuart (1962). "The numerical solution of variational problems". Journal
of Mathematical Analysis and Applications. 5 (1): 30�45. doi:10.1016/0022-
247x(62)90004-5.
Russell, Stuart J.; Norvig, Peter (2010). Artificial Intelligence A Modern
Approach. Prentice Hall. p. 578. ISBN 978-0-13-604259-4. The most popular method
for learning in multilayer networks is called Back-propagation.
Bryson, Arthur Earl (1969). Applied Optimal Control: Optimization, Estimation and
Control. Blaisdell Publishing Company or Xerox College Publishing. p. 481.
Seppo Linnainmaa (1970). The representation of the cumulative rounding error of an
algorithm as a Taylor expansion of the local rounding errors. Master's Thesis (in
Finnish), Univ. Helsinki, 6�7.
Linnainmaa, Seppo (1976). "Taylor expansion of the accumulated rounding error".
BIT Numerical Mathematics. 16 (2): 146�160. doi:10.1007/bf01931367.
Griewank, Andreas (2012). "Who Invented the Reverse Mode of Differentiation?"
(PDF). Documenta Matematica, Extra Volume ISMP: 389�400. Archived from the original
(PDF) on 21 July 2017. Retrieved 27 June 2017.
Griewank, Andreas; Walther, Andrea (2008). Evaluating Derivatives: Principles and
Techniques of Algorithmic Differentiation, Second Edition. SIAM. ISBN 978-0-89871-
776-1.
Dreyfus, Stuart (1973). "The computational solution of optimal control problems
with time lag". IEEE Transactions on Automatic Control. 18 (4): 383�385.
doi:10.1109/tac.1973.1100330.
Paul Werbos (1974). Beyond regression: New tools for prediction and analysis in
the behavioral sciences. PhD thesis, Harvard University.
Werbos, Paul (1982). "Applications of advances in nonlinear sensitivity analysis"
(PDF). System modeling and optimization. Springer. pp. 762�770.
Rumelhart, David E.; Hinton, Geoffrey E.; Williams, Ronald J. (1986). "Learning
representations by back-propagating errors". Nature. 323 (6088): 533�536.
Bibcode:1986Natur.323..533R. doi:10.1038/323533a0.
 Eric A. Wan (1993). "Time series prediction by using a connectionist network with
internal delay lines." In Proceedings of the Santa Fe Institute Studies in the
Sciences of Complexity, 15: p. 195. Addison-Wesley Publishing Co.
Hinton, G.; Deng, L.; Yu, D.; Dahl, G. E.; Mohamed, A. r; Jaitly, N.; Senior, A.;
Vanhoucke, V.; Nguyen, P. (November 2012). "Deep Neural Networks for Acoustic
Modeling in Speech Recognition: The Shared Views of Four Research Groups". IEEE
Signal Processing Magazine. 29 (6): 82�97. Bibcode:2012ISPM...29...82H.
doi:10.1109/msp.2012.2205597. ISSN 1053-5888.
Huang, Guang-Bin; Zhu, Qin-Yu; Siew, Chee-Kheong (2006). "Extreme learning
machine: theory and applications". Neurocomputing. 70 (1): 489�501. CiteSeerX
10.1.1.217.3692. doi:10.1016/j.neucom.2005.12.126.
Widrow, Bernard; et al. (2013). "The no-prop algorithm: A new learning algorithm
for multilayer neural networks". Neural Networks. 37: 182�188.
doi:10.1016/j.neunet.2012.09.020. PMID 23140797.
Ollivier, Yann; Charpiat, Guillaume (2015). "Training recurrent networks without
backtracking". arXiv:1507.07680 [cs.NE].
ESANN. 2009
Hinton, G. E. (2010). "A Practical Guide to Training Restricted Boltzmann
Machines". Tech. Rep. UTML TR 2010-003.
Ojha, Varun Kumar; Abraham, Ajith; Sn�el, V�clav (1 April 2017). "Metaheuristic
design of feedforward neural networks: A review of two decades of research".
Engineering Applications of Artificial Intelligence. 60: 97�116. arXiv:1705.05584.
Bibcode:2017arXiv170505584O. doi:10.1016/j.engappai.2017.01.013.
Dominic, S.; Das, R.; Whitley, D.; Anderson, C. (July 1991). "Genetic
reinforcement learning for neural networks". IJCNN-91-Seattle International Joint
Conference on Neural Networks. IJCNN-91-Seattle International Joint Conference on
Neural Networks. Seattle, Washington, USA: IEEE. doi:10.1109/IJCNN.1991.155315.
ISBN 0-7803-0164-1.
Hoskins, J.C.; Himmelblau, D.M. (1992). "Process control via artificial neural
networks and reinforcement learning". Computers & Chemical Engineering. 16 (4):
241�251. doi:10.1016/0098-1354(92)80045-B.
Bertsekas, D.P.; Tsitsiklis, J.N. (1996). Neuro-dynamic programming. Athena
Scientific. p. 512. ISBN 978-1-886529-10-6.
Secomandi, Nicola (2000). "Comparing neuro-dynamic programming algorithms for the
vehicle routing problem with stochastic demands". Computers & Operations Research.
27 (11�12): 1201�1225. CiteSeerX 10.1.1.392.4034. doi:10.1016/S0305-0548(99)00146-
X.
de Rigo, D.; Rizzoli, A. E.; Soncini-Sessa, R.; Weber, E.; Zenesi, P. (2001).
"Neuro-dynamic programming for the efficient management of reservoir networks".
Proceedings of MODSIM 2001, International Congress on Modelling and Simulation.
MODSIM 2001, International Congress on Modelling and Simulation. Canberra,
Australia: Modelling and Simulation Society of Australia and New Zealand.
doi:10.5281/zenodo.7481. ISBN 0-867405252.
Damas, M.; Salmeron, M.; Diaz, A.; Ortega, J.; Prieto, A.; Olivares, G. (2000).
"Genetic algorithms and neuro-dynamic programming: application to water supply
networks". Proceedings of 2000 Congress on Evolutionary Computation. 2000 Congress
on Evolutionary Computation. La Jolla, California, USA: IEEE.
doi:10.1109/CEC.2000.870269. ISBN 0-7803-6375-2.
Deng, Geng; Ferris, M.C. (2008). Neuro-dynamic programming for fractionated
radiotherapy planning. Springer Optimization and Its Applications. 12. pp. 47�70.
CiteSeerX 10.1.1.137.8288. doi:10.1007/978-0-387-73299-2_3. ISBN 978-0-387-73298-5.
M. Forouzanfar; H. R. Dajani; V. Z. Groza; M. Bolic & S. Rajan (July 2010).
Comparison of Feed-Forward Neural Network Training Algorithms for Oscillometric
Blood Pressure Estimation. 4th Int. Workshop Soft Computing Applications. Arad,
Romania: IEEE.
de Rigo, D.; Castelletti, A.; Rizzoli, A. E.; Soncini-Sessa, R.; Weber, E.
(January 2005). "A selective improvement technique for fastening Neuro-Dynamic
Programming in Water Resources Network Management". In Pavel Z�tek (ed.).
Proceedings of the 16th IFAC World Congress � IFAC-PapersOnLine. 16th IFAC World
Congress. 16. Prague, Czech Republic: IFAC. doi:10.3182/20050703-6-CZ-1902.02172.
ISBN 978-3-902661-75-3. Retrieved 30 December 2011.
Ferreira, C. (2006). "Designing Neural Networks Using Gene Expression Programming"
(PDF). In A. Abraham, B. de Baets, M. K�ppen, and B. Nickolay, eds., Applied Soft
Computing Technologies: The Challenge of Complexity, pages 517�536, Springer-
Verlag.
Da, Y.; Xiurun, G. (July 2005). T. Villmann (ed.). An improved PSO-based ANN with
simulated annealing technique. New Aspects in Neurocomputing: 11th European
Symposium on Artificial Neural Networks. Elsevier.
doi:10.1016/j.neucom.2004.07.002.
Wu, J.; Chen, E. (May 2009). Wang, H.; Shen, Y.; Huang, T.; Zeng, Z. (eds.). A
Novel Nonparametric Regression Ensemble for Rainfall Forecasting Using Particle
Swarm Optimization Technique Coupled with Artificial Neural Network. 6th
International Symposium on Neural Networks, ISNN 2009. Springer. doi:10.1007/978-3-
642-01513-7-6. ISBN 978-3-642-01215-0.
Ting Qin, et al. "A learning algorithm of CMAC based on RLS." Neural Processing
Letters 19.1 (2004): 49�61.
Ting Qin, et al. "Continuous CMAC-QRLS and its systolic array." Neural Processing
Letters 22.1 (2005): 1�16.
Werbos, Paul J. (1994). The Roots of Backpropagation. From Ordered Derivatives to
Neural Networks and Political Forecasting. New York, NY: John Wiley & Sons, Inc.
Li, Y.; Fu, Y.; Li, H.; Zhang, S. W. (1 June 2009). The Improved Training
Algorithm of Back Propagation Neural Network with Self-adaptive Learning Rate. 2009
International Conference on Computational Intelligence and Natural Computing. 1.
pp. 73�76. doi:10.1109/CINC.2009.111. ISBN 978-0-7695-3645-3.
Ivakhnenko, Alexey Grigorevich (1968). "The group method of data handling � a
rival of the method of stochastic approximation". Soviet Automatic Control. 13 (3):
43�55.
Ivakhnenko, Alexey (1971). "Polynomial theory of complex systems". IEEE
Transactions on Systems, Man and Cybernetics (4) (4): 364�378.
doi:10.1109/TSMC.1971.4308320.
Kondo, T.; Ueno, J. (2008). "Multi-layered GMDH-type neural network self-selecting
optimum neural network architecture and its application to 3-dimensional medical
image recognition of blood vessels". International Journal of Innovative Computing,
Information and Control. 4 (1): 175�187.
Fukushima, K. (1980). "Neocognitron: A self-organizing neural network model for a
mechanism of pattern recognition unaffected by shift in position". Biol. Cybern. 36
(4): 193�202. doi:10.1007/bf00344251. PMID 7370364.
LeCun et al., "Backpropagation Applied to Handwritten Zip Code Recognition,"
Neural Computation, 1, pp. 541�551, 1989.
Yann LeCun (2016). Slides on Deep Learning Online
"Unsupervised Feature Learning and Deep Learning Tutorial".
Szegedy, Christian; Liu, Wei; Jia, Yangqing; Sermanet, Pierre; Reed, Scott;
Anguelov, Dragomir; Erhan, Dumitru; Vanhoucke, Vincent; Rabinovich, Andrew (2014).
Going Deeper with Convolutions. Computing Research Repository. p. 1.
arXiv:1409.4842. doi:10.1109/CVPR.2015.7298594. ISBN 978-1-4673-6964-0.
Ran, Lingyan; Zhang, Yanning; Zhang, Qilin; Yang, Tao (12 June 2017).
"Convolutional Neural Network-Based Robot Navigation Using Uncalibrated Spherical
Images" (PDF). Sensors. 17 (6): 1341. doi:10.3390/s17061341. ISSN 1424-8220. PMC
5492478. PMID 28604624.
Hinton, Geoffrey E.; Krizhevsky, Alex; Wang, Sida D. (2011), "Transforming Auto-
Encoders", Lecture Notes in Computer Science, Springer Berlin Heidelberg, pp.
44�51, CiteSeerX 10.1.1.220.5099, doi:10.1007/978-3-642-21735-7_6, ISBN
9783642217340
Hochreiter, Sepp; Schmidhuber, J�rgen (1 November 1997). "Long Short-Term Memory".
Neural Computation. 9 (8): 1735�1780. doi:10.1162/neco.1997.9.8.1735. ISSN 0899-
7667.
"Learning Precise Timing with LSTM Recurrent Networks (PDF Download Available)".
Crossref Listing of Deleted Dois. 1: 115�143. 2000. doi:10.1162/153244303768966139.
Retrieved 13 June 2017.
Bayer, Justin; Wierstra, Daan; Togelius, Julian; Schmidhuber, J�rgen (14 September
2009). Evolving Memory Cell Structures for Sequence Learning (PDF). Artificial
Neural Networks � ICANN 2009. Lecture Notes in Computer Science. 5769. Springer,
Berlin, Heidelberg. pp. 755�764. doi:10.1007/978-3-642-04277-5_76. ISBN 978-3-642-
04276-8.
Fern�ndez, Santiago; Graves, Alex; Schmidhuber, J�rgen (2007). "Sequence labelling
in structured domains with hierarchical recurrent neural networks". In Proc. 20th
Int. Joint Conf. On Artificial In?ligence, Ijcai 2007: 774�779. CiteSeerX
10.1.1.79.1887.
Graves, Alex; Fern�ndez, Santiago; Gomez, Faustino (2006). "Connectionist temporal
classification: Labelling unsegmented sequence data with recurrent neural
networks". In Proceedings of the International Conference on Machine Learning, ICML
2006: 369�376. CiteSeerX 10.1.1.75.6306.
Graves, Alex; Eck, Douglas; Beringer, Nicole; Schmidhuber, J�rgen (2003).
"Biologically Plausible Speech Recognition with LSTM Neural Nets" (PDF). 1st Intl.
Workshop on Biologically Inspired Approaches to Advanced Information Technology,
Bio-ADIT 2004, Lausanne, Switzerland. pp. 175�184.
Fern�ndez, Santiago; Graves, Alex; Schmidhuber, J�rgen (2007). An Application of
Recurrent Neural Networks to Discriminative Keyword Spotting. Proceedings of the
17th International Conference on Artificial Neural Networks. ICANN'07. Berlin,
Heidelberg: Springer-Verlag. pp. 220�229. ISBN 978-3540746935.
Hannun, Awni; Case, Carl; Casper, Jared; Catanzaro, Bryan; Diamos, Greg; Elsen,
Erich; Prenger, Ryan; Satheesh, Sanjeev; Sengupta, Shubho (17 December 2014). "Deep
Speech: Scaling up end-to-end speech recognition". arXiv:1412.5567 [cs.CL].
Sak, Hasim; Senior, Andrew; Beaufays, Francoise (2014). "Long Short-Term Memory
recurrent neural network architectures for large scale acoustic modeling" (PDF).
Li, Xiangang; Wu, Xihong (15 October 2014). "Constructing Long Short-Term Memory
based Deep Recurrent Neural Networks for Large Vocabulary Speech Recognition".
arXiv:1410.4281 [cs.CL].
Fan, Y.; Qian, Y.; Xie, F.; Soong, F. K. (2014). "TTS synthesis with bidirectional
LSTM based Recurrent Neural Networks". Proceedings of the Annual Conference of the
International Speech Communication Association, Interspeech: 1964�1968. Retrieved
13 June 2017.
Zen, Heiga; Sak, Hasim (2015). "Unidirectional Long Short-Term Memory Recurrent
Neural Network with Recurrent Output Layer for Low-Latency Speech Synthesis" (PDF).
Google.com. ICASSP. pp. 4470�4474.
Fan, Bo; Wang, Lijuan; Soong, Frank K.; Xie, Lei (2015). "Photo-Real Talking Head
with Deep Bidirectional LSTM" (PDF). Proceedings of ICASSP.
Sak, Hasim; Senior, Andrew; Rao, Kanishka; Beaufays, Fran�oise; Schalkwyk, Johan
(September 2015). "Google voice search: faster and more accurate".
Gers, Felix A.; Schmidhuber, J�rgen (2001). "LSTM Recurrent Networks Learn Simple
Context Free and Context Sensitive Languages". IEEE Transactions on Neural
Networks. 12 (6): 1333�1340. doi:10.1109/72.963769. PMID 18249962.
Schmidhuber, Juergen (2018). "Video-based Sign Language Recognition without
Temporal Segmentation". arXiv:1801.10111 [cs.CV].
Sutskever, L.; Vinyals, O.; Le, Q. (2014). "Sequence to Sequence Learning with
Neural Networks" (PDF). NIPS'14 Proceedings of the 27th International Conference on
Neural Information Processing Systems. 2: 3104�3112. arXiv:1409.3215.
Bibcode:2014arXiv1409.3215S.
Jozefowicz, Rafal; Vinyals, Oriol; Schuster, Mike; Shazeer, Noam; Wu, Yonghui (7
February 2016). "Exploring the Limits of Language Modeling". arXiv:1602.02410
[cs.CL].
Gillick, Dan; Brunk, Cliff; Vinyals, Oriol; Subramanya, Amarnag (30 November
2015). "Multilingual Language Processing From Bytes". arXiv:1512.00103 [cs.CL].
Vinyals, Oriol; Toshev, Alexander; Bengio, Samy; Erhan, Dumitru (17 November
2014). "Show and Tell: A Neural Image Caption Generator". arXiv:1411.4555 [cs.CV].
Gallicchio, Claudio; Micheli, Alessio; Pedrelli, Luca (2017). "Deep reservoir
computing: A critical experimental analysis". Neurocomputing. 268: 87�99.
doi:10.1016/j.neucom.2016.12.089.
Gallicchio, Claudio; Micheli, Alessio (2017). "Echo State Property of Deep
Reservoir Computing Networks". Cognitive Computation. 9 (3): 337�350.
doi:10.1007/s12559-017-9461-9. ISSN 1866-9956.
Hinton, G.E. (2009). "Deep belief networks". Scholarpedia. 4 (5): 5947.
Bibcode:2009SchpJ...4.5947H. doi:10.4249/scholarpedia.5947.
Larochelle, Hugo; Erhan, Dumitru; Courville, Aaron; Bergstra, James; Bengio,
Yoshua (2007). An Empirical Evaluation of Deep Architectures on Problems with Many
Factors of Variation. Proceedings of the 24th International Conference on Machine
Learning. ICML '07. New York, NY, USA: ACM. pp. 473�480. CiteSeerX 10.1.1.77.3242.
doi:10.1145/1273496.1273556. ISBN 9781595937933.
Graupe, Daniel (2013). Principles of Artificial Neural Networks. World Scientific.
pp. 1�. ISBN 978-981-4522-74-8.
A US 5920852 A D. Graupe," Large memory storage and retrieval (LAMSTAR) network,
April 1996
D. Graupe, "Principles of Artificial Neural Networks.3rd Edition", World
Scientific Publishers, 2013, pp. 203�274.
Nigam, Vivek Prakash; Graupe, Daniel (1 January 2004). "A neural-network-based
detection of epilepsy". Neurological Research. 26 (1): 55�60.
doi:10.1179/016164104773026534. ISSN 0161-6412. PMID 14977058.
Waxman, Jonathan A.; Graupe, Daniel; Carley, David W. (1 April 2010). "Automated
Prediction of Apnea and Hypopnea, Using a LAMSTAR Artificial Neural Network".
American Journal of Respiratory and Critical Care Medicine. 181 (7): 727�733.
doi:10.1164/rccm.200907-1146oc. ISSN 1073-449X. PMID 20019342.
Graupe, D.; Graupe, M. H.; Zhong, Y.; Jackson, R. K. (2008). "Blind adaptive
filtering for non-invasive extraction of the fetal electrocardiogram and its non-
stationarities". Proc. Inst. Mech. Eng. H. 222 (8): 1221�1234.
doi:10.1243/09544119jeim417. PMID 19143416.
Graupe 2013, pp. 240�253
Graupe, D.; Abon, J. (2002). "A Neural Network for Blind Adaptive Filtering of
Unknown Noise from Speech". Intelligent Engineering Systems Through Artificial
Neural Networks. 12: 683�688. Retrieved 14 June 2017.
D. Graupe, "Principles of Artificial Neural Networks.3rd Edition", World
Scientific Publishers", 2013, pp. 253�274.
Girado, J. I.; Sandin, D. J.; DeFanti, T. A. (2003). "Real-time camera-based face
detection using a modified LAMSTAR neural network system". Proc. SPIE 5015,
Applications of Artificial Neural Networks in Image Processing VIII. Applications
of Artificial Neural Networks in Image Processing VIII. 5015: 36�46.
Bibcode:2003SPIE.5015...36G. doi:10.1117/12.477405.
Venkatachalam, V; Selvan, S. (2007). "Intrusion Detection using an Improved
Competitive Learning Lamstar Network". International Journal of Computer Science
and Network Security. 7 (2): 255�263.
Graupe, D.; Smollack, M. (2007). "Control of unstable nonlinear and nonstationary
systems using LAMSTAR neural networks". ResearchGate. Proceedings of 10th IASTED on
Intelligent Control, Sect.592. pp. 141�144. Retrieved 14 June 2017.
Graupe, Daniel (7 July 2016). Deep Learning Neural Networks: Design and Case
Studies. World Scientific Publishing Co Inc. pp. 57�110. ISBN 978-981-314-647-1.
Graupe, D.; Kordylewski, H. (August 1996). Network based on SOM (Self-Organizing-
Map) modules combined with statistical decision tools. Proceedings of the 39th
Midwest Symposium on Circuits and Systems. 1. pp. 471�474 vol.1.
doi:10.1109/mwscas.1996.594203. ISBN 978-0-7803-3636-0.
Graupe, D.; Kordylewski, H. (1 March 1998). "A Large Memory Storage and Retrieval
Neural Network for Adaptive Retrieval and Diagnosis". International Journal of
Software Engineering and Knowledge Engineering. 08 (1): 115�138.
doi:10.1142/s0218194098000091. ISSN 0218-1940.
Kordylewski, H.; Graupe, D; Liu, K. (2001). "A novel large-memory neural network
as an aid in medical diagnosis applications". IEEE Transactions on Information
Technology in Biomedicine. 5 (3): 202�209. doi:10.1109/4233.945291.
Schneider, N.C.; Graupe (2008). "A modified LAMSTAR neural network and its
applications". International Journal of Neural Systems. 18 (4): 331�337.
doi:10.1142/s0129065708001634. PMID 18763732.
Graupe 2013, p. 217
Vincent, Pascal; Larochelle, Hugo; Lajoie, Isabelle; Bengio, Yoshua; Manzagol,
Pierre-Antoine (2010). "Stacked Denoising Autoencoders: Learning Useful
Representations in a Deep Network with a Local Denoising Criterion". The Journal of
Machine Learning Research. 11: 3371�3408.
Ballard, Dana H. (1987). "Modular learning in neural networks" (PDF). Proceedings
of AAAI. pp. 279�284. Archived from the original (PDF) on 16 October 2015.
Deng, Li; Yu, Dong; Platt, John (2012). "Scalable stacking and learning for
building deep architectures" (PDF). 2012 IEEE International Conference on
Acoustics, Speech and Signal Processing (ICASSP): 2133�2136.
Deng, Li; Yu, Dong (2011). "Deep Convex Net: A Scalable Architecture for Speech
Pattern Classification" (PDF). Proceedings of the Interspeech: 2285�2288.
David, Wolpert (1992). "Stacked generalization". Neural Networks. 5 (2): 241�259.
CiteSeerX 10.1.1.133.8090. doi:10.1016/S0893-6080(05)80023-1.
Bengio, Y. (15 November 2009). "Learning Deep Architectures for AI". Foundations
and Trends in Machine Learning. 2 (1): 1�127. CiteSeerX 10.1.1.701.9550.
doi:10.1561/2200000006. ISSN 1935-8237.
Hutchinson, Brian; Deng, Li; Yu, Dong (2012). "Tensor deep stacking networks".
IEEE Transactions on Pattern Analysis and Machine Intelligence. 1�15 (8):
1944�1957. doi:10.1109/tpami.2012.268.
Hinton, Geoffrey; Salakhutdinov, Ruslan (2006). "Reducing the Dimensionality of
Data with Neural Networks". Science. 313 (5786): 504�507.
Bibcode:2006Sci...313..504H. doi:10.1126/science.1127647. PMID 16873662.
Dahl, G.; Yu, D.; Deng, L.; Acero, A. (2012). "Context-Dependent Pre-Trained Deep
Neural Networks for Large-Vocabulary Speech Recognition". IEEE Transactions on
Audio, Speech, and Language Processing. 20 (1): 30�42. CiteSeerX 10.1.1.227.8990.
doi:10.1109/tasl.2011.2134090.
Mohamed, Abdel-rahman; Dahl, George; Hinton, Geoffrey (2012). "Acoustic Modeling
Using Deep Belief Networks". IEEE Transactions on Audio, Speech, and Language
Processing. 20 (1): 14�22. CiteSeerX 10.1.1.338.2670.
doi:10.1109/tasl.2011.2109382.
Courville, Aaron; Bergstra, James; Bengio, Yoshua (2011). "A Spike and Slab
Restricted Boltzmann Machine" (PDF). JMLR: Workshop and Conference Proceeding. 15:
233�241.
Courville, Aaron; Bergstra, James; Bengio, Yoshua (2011). "Unsupervised Models of
Images by Spike-and-Slab RBMs" (PDF). Proceedings of the 28th International
Conference on Machine Learning. 10. pp. 1�8.
Mitchell, T; Beauchamp, J (1988). "Bayesian Variable Selection in Linear
Regression". Journal of the American Statistical Association. 83 (404): 1023�1032.
doi:10.1080/01621459.1988.10478694.
Hinton, Geoffrey; Salakhutdinov, Ruslan (2009). "Efficient Learning of Deep
Boltzmann Machines" (PDF). 3: 448�455.
Larochelle, Hugo; Bengio, Yoshua; Louradour, Jerdme; Lamblin, Pascal (2009).
"Exploring Strategies for Training Deep Neural Networks". The Journal of Machine
Learning Research. 10: 1�40.
Coates, Adam; Carpenter, Blake (2011). "Text Detection and Character Recognition
in Scene Images with Unsupervised Feature Learning" (PDF): 440�445.
Lee, Honglak; Grosse, Roger (2009). Convolutional deep belief networks for
scalable unsupervised learning of hierarchical representations. Proceedings of the
26th Annual International Conference on Machine Learning. pp. 1�8. CiteSeerX
10.1.1.149.6800. doi:10.1145/1553374.1553453. ISBN 9781605585161.
Lin, Yuanqing; Zhang, Tong (2010). "Deep Coding Network" (PDF). Advances in Neural
. . .: 1�9.
Ranzato, Marc Aurelio; Boureau, Y-Lan (2007). "Sparse Feature Learning for Deep
Belief Networks" (PDF). Advances in Neural Information Processing Systems. 23: 1�8.
Socher, Richard; Lin, Clif (2011). "Parsing Natural Scenes and Natural Language
with Recursive Neural Networks" (PDF). Proceedings of the 26th International
Conference on Machine Learning.
Taylor, Graham; Hinton, Geoffrey (2006). "Modeling Human Motion Using Binary
Latent Variables" (PDF). Advances in Neural Information Processing Systems.
Vincent, Pascal; Larochelle, Hugo (2008). Extracting and composing robust features
with denoising autoencoders. Proceedings of the 25th International Conference on
Machine Learning � ICML '08. pp. 1096�1103. CiteSeerX 10.1.1.298.4083.
doi:10.1145/1390156.1390294. ISBN 9781605582054.
Kemp, Charles; Perfors, Amy; Tenenbaum, Joshua (2007). "Learning overhypotheses
with hierarchical Bayesian models". Developmental Science. 10 (3): 307�21.
CiteSeerX 10.1.1.141.5560. doi:10.1111/j.1467-7687.2007.00585.x. PMID 17444972.
Xu, Fei; Tenenbaum, Joshua (2007). "Word learning as Bayesian inference". Psychol.
Rev. 114 (2): 245�72. CiteSeerX 10.1.1.57.9649. doi:10.1037/0033-295X.114.2.245.
PMID 17500627.
Chen, Bo; Polatkan, Gungor (2011). "The Hierarchical Beta Process for
Convolutional Factor Analysis and Deep Learning" (PDF). Proceedings of the 28th
International Conference on International Conference on Machine Learning.
Omnipress. pp. 361�368. ISBN 978-1-4503-0619-5.
Fei-Fei, Li; Fergus, Rob (2006). "One-shot learning of object categories". IEEE
Transactions on Pattern Analysis and Machine Intelligence. 28 (4): 594�611.
CiteSeerX 10.1.1.110.9024. doi:10.1109/TPAMI.2006.79. PMID 16566508.
Rodriguez, Abel; Dunson, David (2008). "The Nested Dirichlet Process". Journal of
the American Statistical Association. 103 (483): 1131�1154. CiteSeerX
10.1.1.70.9873. doi:10.1198/016214508000000553.
Ruslan, Salakhutdinov; Joshua, Tenenbaum (2012). "Learning with Hierarchical-Deep
Models". IEEE Transactions on Pattern Analysis and Machine Intelligence. 35 (8):
1958�71. CiteSeerX 10.1.1.372.909. doi:10.1109/TPAMI.2012.269. PMID 23787346.
Chalasani, Rakesh; Principe, Jose (2013). "Deep Predictive Coding Networks".
arXiv:1301.3541 [cs.LG].
Hinton, Geoffrey E. (1984). "Distributed representations". Archived from the
original on 2 May 2016.
S. Das, C.L. Giles, G.Z. Sun, "Learning Context Free Grammars: Limitations of a
Recurrent Neural Network with an External Stack Memory," Proc. 14th Annual Conf. of
the Cog. Sci. Soc., p. 79, 1992.
Mozer, M. C.; Das, S. (1993). A connectionist symbol manipulator that discovers
the structure of context-free languages. NIPS 5. pp. 863�870.
Schmidhuber, J. (1992). "Learning to control fast-weight memories: An alternative
to recurrent nets". Neural Computation. 4 (1): 131�139.
doi:10.1162/neco.1992.4.1.131.
Gers, F.; Schraudolph, N.; Schmidhuber, J. (2002). "Learning precise timing with
LSTM recurrent networks" (PDF). JMLR. 3: 115�143.
J�rgen Schmidhuber (1993). "An introspective network that can learn to run its own
weight change algorithm". In Proc. of the Intl. Conf. on Artificial Neural
Networks, Brighton. IEE. pp. 191�195.
Hochreiter, Sepp; Younger, A. Steven; Conwell, Peter R. (2001). "Learning to Learn
Using Gradient Descent". ICANN. 2130: 87�94. CiteSeerX 10.1.1.5.323.
Schmidhuber, Juergen (2015). "Learning to Transduce with Unbounded Memory".
arXiv:1506.02516 [cs.NE].
Schmidhuber, Juergen (2014). "Neural Turing Machines". arXiv:1410.5401 [cs.NE].
Burgess, Matt. "DeepMind's AI learned to ride the London Underground using human-
like reason and memory". WIRED UK. Retrieved 19 October 2016.
"DeepMind AI 'Learns' to Navigate London Tube". PCMAG. Retrieved 19 October 2016.
Mannes, John. "DeepMind's differentiable neural computer helps you navigate the
subway with its memory". TechCrunch. Retrieved 19 October 2016.
Graves, Alex; Wayne, Greg; Reynolds, Malcolm; Harley, Tim; Danihelka, Ivo;
Grabska-Barwinska, Agnieszka; Colmenarejo, Sergio G�mez; Grefenstette, Edward;
Ramalho, Tiago (12 October 2016). "Hybrid computing using a neural network with
dynamic external memory". Nature. 538 (7626): 471�476. Bibcode:2016Natur.538..471G.
doi:10.1038/nature20101. ISSN 1476-4687. PMID 27732574.
"Differentiable neural computers | DeepMind". DeepMind. Retrieved 19 October 2016.
 Atkeson, Christopher G.; Schaal, Stefan (1995). "Memory-based neural networks for
robot learning". Neurocomputing. 9 (3): 243�269. doi:10.1016/0925-2312(95)00033-6.
Salakhutdinov, Ruslan, and Geoffrey Hinton. "Semantic hashing." International
Journal of Approximate Reasoning 50.7 (2009): 969�978.
Le, Quoc V.; Mikolov, Tomas (2014). "Distributed representations of sentences and
documents". arXiv:1405.4053 [cs.CL].
Schmidhuber, Juergen (2014). "Memory Networks". arXiv:1410.3916 [cs.AI].
Schmidhuber, Juergen (2015). "End-To-End Memory Networks". arXiv:1503.08895
[cs.NE].
Schmidhuber, Juergen (2015). "Large-scale Simple Question Answering with Memory
Networks". arXiv:1506.02075 [cs.LG].
"AI device identifies objects at the speed of light: The 3D-printed artificial
neural network can be used in medicine, robotics and security". ScienceDaily.
Retrieved 8 August 2018.
Schmidhuber, Juergen (2015). "Pointer Networks". arXiv:1506.03134 [stat.ML].
Schmidhuber, Juergen (2015). "Neural Random-Access Machines". arXiv:1511.06392
[cs.LG].
Kalchbrenner, N.; Blunsom, P. (2013). "Recurrent continuous translation models".
EMNLP'2013.
Sutskever, I.; Vinyals, O.; Le, Q. V. (2014). "Sequence to sequence learning with
neural networks" (PDF). NIPS'2014. arXiv:1409.3215 [cs.CL]. Unknown parameter |
publisher= ignored (help)
Schmidhuber, Juergen (2014). "Learning Phrase Representations using RNN Encoder-
Decoder for Statistical Machine Translation". arXiv:1406.1078 [cs.CL].
Schmidhuber, Juergen; Courville, Aaron; Bengio, Yoshua (2015). "Describing
Multimedia Content using Attention-based Encoder--Decoder Networks". IEEE
Transactions on Multimedia. 17 (11): 1875�1886. arXiv:1507.01053.
Bibcode:2015arXiv150701053C. doi:10.1109/TMM.2015.2477044.
Scholkopf, B; Smola, Alexander (1998). "Nonlinear component analysis as a kernel
eigenvalue problem". Neural Computation. (44) (5): 1299�1319. CiteSeerX
10.1.1.53.8911. doi:10.1162/089976698300017467.
Cho, Youngmin (2012). "Kernel Methods for Deep Learning" (PDF): 1�9.
Deng, Li; Tur, Gokhan; He, Xiaodong; Hakkani-T�r, Dilek (1 December 2012). "Use of
Kernel Deep Convex Networks and End-To-End Learning for Spoken Language
Understanding". Microsoft Research.
Zoph, Barret; Le, Quoc V. (4 November 2016). "Neural Architecture Search with
Reinforcement Learning". arXiv:1611.01578 [cs.LG].
Zissis, Dimitrios (October 2015). "A cloud based architecture capable of
perceiving and predicting multiple vessel behaviour". Applied Soft Computing. 35:
652�661. doi:10.1016/j.asoc.2015.07.002.
Roman M. Balabin; Ekaterina I. Lomakina (2009). "Neural network approach to
quantum-chemistry data: Accurate prediction of density functional theory energies".
J. Chem. Phys. 131 (7): 074104. Bibcode:2009JChPh.131g4104B. doi:10.1063/1.3206326.
PMID 19708729.
Silver, David, et al. "Mastering the game of Go with deep neural networks and tree
search." nature 529.7587 (2016): 484.
Sengupta, Nandini; Sahidullah, Md; Saha, Goutam (August 2016). "Lung sound
classification using cepstral-based statistical features". Computers in Biology and
Medicine. 75 (1): 118�129. doi:10.1016/j.compbiomed.2016.05.013. PMID 27286184.
Choy, Christopher B., et al. "3d-r2n2: A unified approach for single and multi-
view 3d object reconstruction." European conference on computer vision. Springer,
Cham, 2016.
French, Jordan (2016). "The time traveller's CAPM". Investment Analysts Journal.
46 (2): 81�96. doi:10.1080/10293523.2016.1255469.
Schechner, Sam (15 June 2017). "Facebook Boosts A.I. to Block Terrorist
Propaganda". Wall Street Journal. ISSN 0099-9660. Retrieved 16 June 2017.
Ganesan, N (2010). "Application of Neural Networks in Diagnosing Cancer Disease
Using Demographic Data". International Journal of Computer Applications. 1 (26):
81�97. Bibcode:2010IJCA....1z..81G. doi:10.5120/476-783.
 Bottaci, Leonardo. "Artificial Neural Networks Applied to Outcome Prediction for
Colorectal Cancer Patients in Separate Institutions" (PDF). The Lancet.
Alizadeh, Elaheh; Lyons, Samanthe M; Castle, Jordan M; Prasad, Ashok (2016).
"Measuring systematic changes in invasive cancer cell shape using Zernike moments".
Integrative Biology. 8 (11): 1183�1193. doi:10.1039/C6IB00100A. PMID 27735002.
Lyons, Samanthe (2016). "Changes in cell shape are correlated with metastatic
potential in murine". Biology Open. 5 (3): 289�299. doi:10.1242/bio.013409. PMC
4810736. PMID 26873952.
Nabian, Mohammad Amin; Meidani, Hadi (28 August 2017). "Deep Learning for
Accelerated Reliability Analysis of Infrastructure Networks". Computer-Aided Civil
and Infrastructure Engineering. 33 (6): 443�458. arXiv:1708.08551.
Bibcode:2017arXiv170808551N. doi:10.1111/mice.12359.
Nabian, Mohammad Amin; Meidani, Hadi (2018). "Accelerating Stochastic Assessment
of Post-Earthquake Transportation Network Connectivity via Machine-Learning-Based
Surrogates". Transportation Research Board 97th Annual Meeting.
D�az, E.; Brotons, V.; Tom�s, R. (September 2018). "Use of artificial neural
networks to predict 3-D elastic settlement of foundations on soils with inclined
bedrock". Soils and Foundations. 58 (6): 1414�1422.
doi:10.1016/j.sandf.2018.08.001. hdl:10045/81208. ISSN 0038-0806.
null null (1 April 2000). "Artificial Neural Networks in Hydrology. I: Preliminary
Concepts". Journal of Hydrologic Engineering. 5 (2): 115�123. CiteSeerX
10.1.1.127.3861. doi:10.1061/(ASCE)1084-0699(2000)5:2(115).
null null (1 April 2000). "Artificial Neural Networks in Hydrology. II: Hydrologic
Applications". Journal of Hydrologic Engineering. 5 (2): 124�137. doi:10.1061/
(ASCE)1084-0699(2000)5:2(124).
Peres, D. J.; Iuppa, C.; Cavallaro, L.; Cancelliere, A.; Foti, E. (1 October
2015). "Significant wave height record extension by neural networks and reanalysis
wind data". Ocean Modelling. 94: 128�140. Bibcode:2015OcMod..94..128P.
doi:10.1016/j.ocemod.2015.08.002.
Dwarakish, G. S.; Rakshith, Shetty; Natesan, Usha (2013). "Review on Applications
of Neural Network in Coastal Engineering". Artificial Intelligent Systems and
Machine Learning. 5 (7): 324�331.
Ermini, Leonardo; Catani, Filippo; Casagli, Nicola (1 March 2005). "Artificial
Neural Networks applied to landslide susceptibility assessment". Geomorphology.
Geomorphological hazard and human impact in mountain environments. 66 (1): 327�343.
Bibcode:2005Geomo..66..327E. doi:10.1016/j.geomorph.2004.09.025.
Nix, R.; Zhang, J. (May 2017). "Classification of Android apps and malware using
deep neural networks". 2017 International Joint Conference on Neural Networks
(IJCNN): 1871�1878. doi:10.1109/IJCNN.2017.7966078. ISBN 978-1-5090-6182-2.
"Machine Learning: Detecting malicious domains with Tensorflow". The Coruscan
Project.
"Detecting Malicious URLs". The systems and networking group at UCSD.
"DeepExploit: a fully automated penetration test tool". Isao Takaesu | GitHub. 12
July 2019.
Homayoun, Sajad; Ahmadzadeh, Marzieh; Hashemi, Sattar; Dehghantanha, Ali; Khayami,
Raouf (2018), Dehghantanha, Ali; Conti, Mauro; Dargahi, Tooska (eds.), "BoTShark: A
Deep Learning Approach for Botnet Traffic Detection", Cyber Threat Intelligence,
Advances in Information Security, Springer International Publishing, pp. 137�153,
doi:10.1007/978-3-319-73951-9_7, ISBN 9783319739519
and (January 1994). "Credit card fraud detection with a neural-network". 1994
Proceedings of the Twenty-Seventh Hawaii International Conference on System
Sciences. 3: 621�630. doi:10.1109/HICSS.1994.323314. ISBN 978-0-8186-5090-1.
Nagy, Alexandra (28 June 2019). "Variational Quantum Monte Carlo Method with a
Neural-Network Ansatz for Open Quantum Systems". Physical Review Letters. 122 (25):
250501. arXiv:1902.09483. Bibcode:2019arXiv190209483N.
doi:10.1103/PhysRevLett.122.250501.
Yoshioka, Nobuyuki; Hamazaki, Ryusuke (28 June 2019). "Constructing neural
stationary states for open quantum many-body systems". Physical Review B. 99 (21):
214306. arXiv:1902.07006. Bibcode:2019arXiv190207006Y.
doi:10.1103/PhysRevB.99.214306.
Hartmann, Michael J.; Carleo, Giuseppe (28 June 2019). "Neural-Network Approach to
Dissipative Quantum Many-Body Dynamics". Physical Review Letters. 122 (25): 250502.
arXiv:1902.05131. Bibcode:2019arXiv190205131H. doi:10.1103/PhysRevLett.122.250502.
Vicentini, Filippo; Biella, Alberto; Regnault, Nicolas; Ciuti, Cristiano (28 June
2019). "Variational Neural-Network Ansatz for Steady States in Open Quantum
Systems". Physical Review Letters. 122 (25): 250503. arXiv:1902.10104.
Bibcode:2019arXiv190210104V. doi:10.1103/PhysRevLett.122.250503.
Forrest MD (April 2015). "Simulation of alcohol action upon a detailed Purkinje
neuron model and a simpler surrogate model that runs >400 times faster". BMC
Neuroscience. 16 (27): 27. doi:10.1186/s12868-015-0162-6. PMC 4417229. PMID
25928094.
Siegelmann, H.T.; Sontag, E.D. (1991). "Turing computability with neural nets"
(PDF). Appl. Math. Lett. 4 (6): 77�80. doi:10.1016/0893-9659(91)90080-F.
Balc�zar, Jos� (July 1997). "Computational Power of Neural Networks: A Kolmogorov
Complexity Characterization". IEEE Transactions on Information Theory. 43 (4):
1175�1183. CiteSeerX 10.1.1.411.7782. doi:10.1109/18.605580.
Crick, Francis (1989). "The recent excitement about neural networks". Nature. 337
(6203): 129�132. Bibcode:1989Natur.337..129C. doi:10.1038/337129a0. PMID 2911347.
Adrian, Edward D. (1926). "The impulses produced by sensory nerve endings". The
Journal of Physiology. 61 (1): 49�72. doi:10.1113/jphysiol.1926.sp002273. PMC
1514809. PMID 16993776.
Dewdney, A. K. (1 April 1997). Yes, we have no neutrons: an eye-opening tour
through the twists and turns of bad science. Wiley. p. 82. ISBN 978-0-471-10806-1.
D. J. Felleman and D. C. Van Essen, "Distributed hierarchical processing in the
primate cerebral cortex," Cerebral Cortex, 1, pp. 1�47, 1991.
J. Weng, "Natural and Artificial Intelligence: Introduction to Computational
Brain-Mind," BMI Press, ISBN 978-0985875725, 2012.
Edwards, Chris (25 June 2015). "Growing pains for deep learning". Communications
of the ACM. 58 (7): 14�16. doi:10.1145/2771283.
Schmidhuber, J�rgen (2015). "Deep learning in neural networks: An overview".
Neural Networks. 61: 85�117. arXiv:1404.7828. doi:10.1016/j.neunet.2014.09.003.
PMID 25462637.
"A Survey of FPGA-based Accelerators for Convolutional Neural Networks", NCAA,
2018
Cade Metz (18 May 2016). "Google Built Its Very Own Chips to Power Its AI Bots".
Wired.
NASA � Dryden Flight Research Center � News Room: News Releases: NASA NEURAL
NETWORK PROJECT PASSES MILESTONE. Nasa.gov. Retrieved on 2013-11-20.
"Roger Bridgman's defence of neural networks". Archived from the original on 19
March 2012. Retrieved 12 July 2010.
"Scaling Learning Algorithms towards {AI} � LISA � Publications � Aigaion 2.0".
Sun and Bookman (1990)
Tahmasebi; Hezarkhani (2012). "A hybrid neural networks-fuzzy logic-genetic
algorithm for grade estimation". Computers & Geosciences. 42: 18�27.
Bibcode:2012CG.....42...18T. doi:10.1016/j.cageo.2012.02.004. PMC 4268588. PMID
25540468.
Bibliography
Bhadeshia H. K. D. H. (1999). "Neural Networks in Materials Science" (PDF). ISIJ
International. 39 (10): 966�979. doi:10.2355/isijinternational.39.966.
M., Bishop, Christopher (1995). Neural networks for pattern recognition. Clarendon
Press. ISBN 978-0198538493. OCLC 33101074.
Borgelt, Christian (2003). Neuro-Fuzzy-Systeme : von den Grundlagen k�nstlicher
Neuronaler Netze zur Kopplung mit Fuzzy-Systemen. Vieweg. ISBN 9783528252656. OCLC
76538146.
Cybenko, G.V. (2006). "Approximation by Superpositions of a Sigmoidal function". In
van Schuppen, Jan H. (ed.). Mathematics of Control, Signals, and Systems. Springer
International. pp. 303�314. PDF
Dewdney, A. K. (1997). Yes, we have no neutrons : an eye-opening tour through the
twists and turns of bad science. New York: Wiley. ISBN 9780471108061. OCLC
35558945.
Duda, Richard O.; Hart, Peter Elliot; Stork, David G. (2001). Pattern
classification (2 ed.). Wiley. ISBN 978-0471056690. OCLC 41347061.
Egmont-Petersen, M.; de Ridder, D.; Handels, H. (2002). "Image processing with
neural networks � a review". Pattern Recognition. 35 (10): 2279�2301. CiteSeerX
10.1.1.21.5444. doi:10.1016/S0031-3203(01)00178-9.
Fahlman, S.; Lebiere, C (1991). "The Cascade-Correlation Learning Architecture"
(PDF).
created for National Science Foundation, Contract Number EET-8716324, and Defense
Advanced Research Projects Agency (DOD), ARPA Order No. 4976 under Contract F33615-
87-C-1499.
Gurney, Kevin (1997). An introduction to neural networks. UCL Press. ISBN 978-
1857286731. OCLC 37875698.
Haykin, Simon S. (1999). Neural networks : a comprehensive foundation. Prentice
Hall. ISBN 978-0132733502. OCLC 38908586.
Hertz, J.; Palmer, Richard G.; Krogh, Anders S. (1991). Introduction to the theory
of neural computation. Addison-Wesley. ISBN 978-0201515602. OCLC 21522159.
Information theory, inference, and learning algorithms. Cambridge University Press.
25 September 2003. Bibcode:2003itil.book.....M. ISBN 9780521642989. OCLC 52377690.
Kruse, Rudolf; Borgelt, Christian; Klawonn, F.; Moewes, Christian; Steinbrecher,
Matthias; Held, Pascal (2013). Computational intelligence : a methodological
introduction. Springer. ISBN 9781447150121. OCLC 837524179.
Lawrence, Jeanette (1994). Introduction to neural networks : design, theory and
applications. California Scientific Software. ISBN 978-1883157005. OCLC 32179420.
MacKay, David, J.C. (2003). Information Theory, Inference, and Learning Algorithms
(PDF). Cambridge University Press. ISBN 9780521642989.
Masters, Timothy (1994). Signal and image processing with neural networks : a C++
sourcebook. J. Wiley. ISBN 978-0471049630. OCLC 29877717.
Ripley, Brian D. (2007). Pattern Recognition and Neural Networks. Cambridge
University Press. ISBN 978-0-521-71770-0.
Siegelmann, H.T.; Sontag, Eduardo D. (1994). "Analog computation via neural
networks" (PDF). Theoretical Computer Science. 131 (2): 331�360. doi:10.1016/0304-
3975(94)90178-3.
Smith, Murray (1993). Neural networks for statistical modeling. Van Nostrand
Reinhold. ISBN 978-0442013103. OCLC 27145760.
Wasserman, Philip D. (1993). Advanced methods in neural computing. Van Nostrand
Reinhold. ISBN 978-0442004613. OCLC 27429729.
External links
The Neural Network Zoo � a compilation of neural network types

vte
Processor technologies
Models
Turing machine Universal Post�Turing Quantum Belt machine Stack machine Finite-
state machine with datapath Hierarchical Queue automaton Register machines Counter
Pointer Random-access Random-access stored program
Architecture
Von Neumann Harvard modified Dataflow Transport-triggered Cellular Endianness
Memory access NUMA HUMA Load/store Register/memory Cache hierarchy Memory hierarchy
Virtual memory Secondary storage Heterogeneous Fabric Multiprocessing Cognitive
Neuromorphic
Instruction set
architectures
Types
CISC RISC Application-specific EDGE TRIPS VLIW EPIC MISC OISC NISC ZISC comparison
addressing modes
x86 ARM MIPS Power ISA SPARC Itanium Unicore MicroBlaze RISC-V others
Execution
Instruction pipelining
Pipeline stall Operand forwarding Classic RISC pipeline
Hazards
Data dependency Structural Control False sharing
Out-of-order
Tomasulo algorithm Reservation station Re-order buffer Register renaming
Speculative
Branch prediction Memory dependence prediction
Parallelism
Level
Bit Bit-serial Word Instruction Pipelining Scalar Superscalar Task Thread Process
Data Vector Memory Distributed
Multithreading
Temporal Simultaneous Hyperthreading Speculative Preemptive Cooperative
Flynn's taxonomy
SISD SIMD SWAR SIMT MISD MIMD SPMD
Processor
performance
Transistor count Instructions per cycle (IPC) Cycles per instruction (CPI)
Instructions per second (IPS) Floating-point operations per second (FLOPS)
Transactions per second (TPS) Synaptic updates per second (SUPS) Performance per
watt (PPW) Cache performance metrics Computer performance by orders of magnitude
Types
Central processing unit (CPU) Graphics processing unit (GPU) GPGPU Vector Barrel
Stream Coprocessor ASIC FPGA CPLD Multi-chip module (MCM) System in package (SiP)
By application
Microprocessor Microcontroller Mobile Notebook Ultra-low-voltage ASIP
Systems
on chip
System on a chip (SoC) Multiprocessor (MPSoC) Programmable (PSoC) Network on a chip
(NoC)
Hardware
accelerators
AI accelerator Vision processing unit (VPU) Physics processing unit (PPU) Digital
signal processor (DSP) Tensor processing unit (TPU) Secure cryptoprocessor Network
processor Baseband processor
Word size
1-bit 2-bit 4-bit 8-bit 16-bit 32-bit 48-bit 64-bit 128-bit 256-bit 512-bit others
variable
Core count
Single-core Multi-core Manycore Heterogeneous architecture
Components
Core Cache CPU cache replacement policies coherence Bus Clock rate Clock signal
FIFO
Functional units
Arithmetic logic unit (ALU) Address generation unit (AGU) Floating-point unit (FPU)
Memory management unit (MMU) Load�store unit Translation lookaside buffer (TLB)
Integrated memory controller (IMC)
Logic
Combinational Sequential Glue Logic gate Quantum Array
Registers
Processor register Status register Stack register Register file Memory buffer
Program counter
Control unit
Instruction unit Data buffer Write buffer Microcode ROM Counter
Datapath
Multiplexer Demultiplexer Adder Multiplier CPU Binary decoder Address decoder Sum
addressed decoder Barrel shifter
Circuitry
Integrated circuit 3D Mixed-signal Power management Boolean Digital Analog Quantum
Switch
Power
management
PMU APM ACPI Dynamic frequency scaling Dynamic voltage scaling Clock gating
Performance per watt (PPW)
Related
History of general-purpose CPUs Microprocessor chronology Processor design Digital
electronics Hardware security module Semiconductor device fabrication
Categories: Computational statisticsArtificial neural networksClassification
algorithmsComputational neuroscienceMarket researchMarket segmentationMathematical
psychologyMathematical and quantitative methods (economics)
Navigation menu
Not logged inTalkContributionsCreate accountLog inArticleTalkReadEditView
historySearch
Search Wikipedia
Main page
Contents
Featured content
Current events
Random article
Donate to Wikipedia
Wikipedia store
Interaction
Help
About Wikipedia
Community portal
Recent changes
Contact page
Tools
What links here
Related changes
Upload file
Special pages
Permanent link
Page information
Wikidata item
Cite this page
In other projects
Wikimedia Commons
Print/export
Create a book
Download as PDF
Printable version

Languages
???????
Espa�ol
??????
Bahasa Indonesia
Bahasa Melayu
Portugu�s
???????
????
??
46 more
Edit links
This page was last edited on 12 July 2019, at 21:17 (UTC).
Text is available under the Creative Commons Attribution-ShareAlike License;
additional terms may apply. By using this site, you agree to the Terms of Use and
Privacy Policy. Wikipedia� is a registered trademark of the Wikimedia Foundation,
Inc., a non-profit organization.
Privacy policyAbout WikipediaDisclaimersContact WikipediaDevelopersCookie
statementMobile viewWikimedia Foundation Powered by MediaWiki