Scribd
Upload
2 views3 pages

Neural Network

Artificial Neural Networks (ANNs) are inspired by the human brain's structure and are used for tasks like forecasting when traditional models are inadequate. They consist of layers that process inputs through nodes with activation functions, and their performance can be affected by issues like overfitting. While ANNs have applications in various fields, they should complement rather than replace traditional statistical methods.

Uploaded by

linren2005
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
2 views3 pages

Neural Network

Artificial Neural Networks (ANNs) are inspired by the human brain's structure and are used for tasks like forecasting when traditional models are inadequate. They consist of layers that process inputs through nodes with activation functions, and their performance can be affected by issues like overfitting. While ANNs have applications in various fields, they should complement rather than replace traditional statistical methods.

Uploaded by

linren2005
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

1.

Introduction to Artificial Neural Networks


●​ Neural networks are inspired by how the human brain processes information.
●​ The brain's basic unit is the neuron, and it contains about 10 billion neurons and around
60 trillion synapses.
●​ These are arranged in a nonlinear, complex, and parallel structure.
●​ This biological setup is extremely efficient for learning, reasoning, and handling large
volumes of data.

2. Purpose and Use in Forecasting


●​ Artificial neural networks (ANNs) are built to mimic human problem-solving capabilities.
●​ They are often referred to as black-box models because they work without needing a
clear mathematical representation of the system.
●​ In forecasting, ANNs are used when analytical models are hard to build.
●​ For example, a business may use a neural network to predict product demand instead of
developing a detailed statistical forecasting model—if the results are accurate and
efficient.

3. Multilayer Feedforward Neural Networks


●​ These are the most commonly used neural networks for prediction and regression tasks.
●​ They model the relationship between multiple predictor variables(x1,x2,.....,xp) and
response variable(s) y .
●​ Inputs may include current explanatory variables and past values of the target variable
(lagged inputs).
●​ The typical structure involves:
○​ Input layer: Original predictor variables.
○​ Hidden layer: Derived internal representations.
○​ Output layer: Final prediction/output values.
4. Working of Nodes and Activation Functions
●​ Each node computes a weighted sum of the inputs from the previous layer and adds a
bias term (like an intercept).
●​ This sum is passed through a nonlinear transformation (activation function).

●​ For the hidden layer node , the output is .


●​ These transformed values are again combined to produce the final output prediction
, where .

5. Choice of Activation Functions


●​ Activation functions introduce non-linearity and help model complex relationships.
●​ Common choices:
○​ Hidden layers: logistic (sigmoid) or hyperbolic tangent (tanh) functions.
○​ Output layer:
■​ Sigmoid for classification (bounded output).
■​ Identity function (no transformation) for regression or continuous
outcomes.

6. Parameter Estimation (Training the Network)


●​ ANNs have many weights (parameters), making them powerful and flexible.
●​ These weights must be estimated using data, a process called training.
●​ Typically, the goal is to minimize the residual sum of squares (difference between
predicted and actual values).
●​ This is a nonlinear optimization problem; common methods include:
○​ Backpropagation: a variation of gradient descent.
○​ Other gradient-based techniques.
●​ Inputs are often standardized, and small random values are used as initial weight values.

7. Overfitting and Its Solutions


●​ A major issue in ANNs is overfitting—performing well on training data but poorly on
unseen (test) data.
●​ Overfitting happens due to excessive model complexity.
●​ Solutions include:
○​ Optimal brain surgery: Pruning unnecessary weights.
○​ Early stopping: Halting training before full convergence.
○​ Regularization: Adding penalties for large weights to the loss function.
○​ Cross-validation: Used to decide training length and architecture.
8. Network Architecture and Model Selection
●​ Network architecture involves:
○​ Number of hidden layers.
○​ Number of neurons in each layer.
○​ Choice of activation functions.
●​ These choices greatly impact performance.
●​ Cross-validation helps in selecting an architecture that generalizes well to new data.

9. Applications and Limitations


●​ ANNs are widely used in many fields:
○​ Time series forecasting
○​ Image recognition
○​ Natural language processing
○​ Financial modeling
●​ However, many practitioners overlook overfitting risks.
●​ Lack of training in statistical modeling can lead to misuse.
●​ Many software tools for neural networks don't address overfitting properly.
●​ Studies often find that neural networks perform poorly on future data not used in training.

10. Conclusion
●​ Neural networks are not a replacement but a complement to traditional statistical
forecasting tools.
●​ They should be used thoughtfully, alongside techniques like regression, ARIMA, and
others, especially where nonlinear patterns are expected.

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy