ARTIFICIAL INTELLIGENCE (AI) IN

SALES MKT-4000 MODULE 5

DEEP DIVE INTO AI

10 Machine Learning Algorithms to Know in 2024
https://www.coursera.org/articles/machine-learning-algorithms
Problems with non-deep ML models

Manual feature engineering, where the data scientist must carefully select
and transform features to improve model performance.

Limited Capacity for Complex Patterns and relationships in the data.

As the size and complexity of datasets increase, some non-deep learning

models may become computationally expensive or inefficient.
Deep Learning
Deep learning is a subset of machine learning that uses multilayered neural networks,
called deep neural networks - human brain.
Deep neural networks include an input layer, at least three but usually hundreds of
hidden layers, and an output layer.
The layered network can process extensive amounts of data and determine the “weight”
of each link in the network — for example, in an image recognition system, some layers
of the neural network might detect individual features of a face, like eyes, nose, or
mouth, while another layer would be able to tell whether those features appear in a way
that indicates a face.

https://www.ibm.com/topics/artificial-intelligence
https://mitsloan.mit.edu/ideas-made-to-matter/machine-learning-explained
Deep Learning

https://www.ibm.com/topics/artificial-intelligence
Machine Learning vs Deep Learning

The chief difference between deep learning and machine

learning is the structure of the underlying neural network
architecture.

“Nondeep,” traditional machine learning models use simple

neural networks with one or two computational layers.
Deep Learning …cont
These multiple layers enable unsupervised learning: they can automate the
extraction of features from large, unlabeled and unstructured data sets, and
make their own predictions about what the data represents.

Because deep learning doesn’t require human intervention, it enables

machine learning at a tremendous scale.
Deep Learning … cont
Suited to natural language processing (NLP), computer vision, and other
tasks that involve fast, accurate identification complex patterns and
relationships in large amounts of data.
Deep learning also enables:
Semi-supervised learning, which combines supervised and unsupervised learning by
using both labeled and unlabeled data.
Self-supervised learning, which generates implicit labels from unstructured data.
Reinforcement learning.
Transfer learning, in which knowledge gained through one task or data set is used to
improve model performance on another related task or different data set.
Types of Deep Learning

Convolutional neural networks (CNNs)

Recurrent neural networks (RNNs)
Autoencoders and variational autoencoders (VAEs)
Generative adversarial networks (GANs)
Diffusion Models
Transformer Models
Convolutional neural networks (CNNs)
Convolutional neural networks (CNNs) are used primarily in computer vision and image
classification applications. They can detect features and patterns within images and
videos, enabling tasks such as object detection, image recognition, pattern recognition and
face recognition.

CNNs are a specific type of neural network, which is composed of node layers, containing
an input layer, one or more hidden layers and an output layer. Each node connects to
another and has an associated weight and threshold.
https://www.linkedin.com/pulse/what-convolutional-neural-network-cnn-deep-learning-nafiz-shahriar/
Recurrent neural networks (RNNs)
RNNs use sequential or time-series data.

RNNs can be identified by their feedback loops.

Predictions about future outcomes.

RNNs use their “memory” as they take information from prior inputs to influence the current
input and output.
RNNs: A Simplified Explanation

https://mohamedbakrey094.medium.com/all-about-recurrent-neural-network-rnn-a236ad5c59f4
RNN Applications
• Language translation: Translating text from one language to another.
• Speech recognition: Converting spoken words into text.
• Generating text: Creating new text, such as poems or articles.
• Time series analysis: Predicting future trends based on past data.
Autoencoders and variational autoencoders (VAEs)
Deep learning - numerical data - analysis of images, speech and other complex data types.

VAEs are used for generating realistic images and speech — cornerstone of what we think
of as generative AI.

Variational autoencoders added the critical ability not just to reconstruct data, but also
output variations on the original data.

Two types:
Regular autoencoders: These try to reconstruct the input data as accurately as possible.
VAEs: These not only try to reconstruct the input data but also generate new, similar data.
How does it work?
1. Encoder: A module that compresses the train-validate-test set input
data into an encoded representation that is typically several orders of
magnitude smaller than the input data.
2. Bottleneck: A module that contains the compressed knowledge
representations.
3. Decoder: A module that helps the network “decompress” the knowledge
representations and reconstructs the data back from its encoded form.
The output is then compared with a ground truth.
Autoencoders & VAEs: A Simplified Explanation

https://www.v7labs.com/blog/autoencoders-guide
 Autoencoders & VAEs - Applications
Image Compression: Reducing the size of images while preserving their
quality.

Anomaly Detection: Identifying unusual or abnormal data points.

--------------------------------------------------------------------------------------------------------
Generating images: Creating new images that look like real photos or
paintings.
Data Augmentation: Generating additional training data to improve model
performance.
Drug discovery: Designing new drugs by generating new molecular structures.
Generative adversarial networks (GANs)
GANs create new data resembling the original training data. For example, images
appearing to be human faces—but are generated, not taken of real people.
The “adversarial” part of the name comes from the back-and-forth between the two
portions of the GAN: a generator and a discriminator.
The generator creates something: images, video or audio and then producing an output
with a twist. For example, a horse can be transformed into a zebra with some degree of
accuracy.
The result depends on the input and how well-trained the layers are in the generative
model for this use case.
The discriminator is the adversary, where the generative result (fake image) is compared
against the real images in the dataset.
The discriminator tries to distinguish between the real and fake images, video or audio.
A Simplified Explanation of GANs

https://developers.google.com/machine-learning/gan/gan_structure
GAN applications
• Creating realistic images: GANs can be used to generate new images that look like real
photos or paintings.
• Generating music: GANs can create new music pieces in different styles.
• Enhancing images: GANs can be used to improve the quality of images or videos.

Limitations:
• Training a GAN can be time-consuming and computationally expensive.
• GANs may generate images or other data that don't make sense in the real world.
Diffusion Models

Diffusion models are generative models that are trained using the forward and reverse
diffusion process of progressive noise-addition and denoising.
They generate data similar to the data on which they are trained, but then overwrite the
data used to train them.
A diffusion model learns to minimize the differences of the generated samples versus the
desired target.
Diffusion models have the advantage of not requiring adversarial training, which speeds
the learning process and offers process control.
 Diffusion Models – How it works
1. Add noise: The model starts by adding random noise to the data. This makes the data look
unrecognizable.
2. Learn to denoise: The model learns to remove the noise and recover the original data.
3. Generate new data: Once the model has learned to denoise the data, it can be used to generate
new, similar data.

https://encord.com/blog/diffusion-models/
Diffusion models advantages over other generative
models

• Diffusion models are generally easier to train than other generative models, such as GANs.
• Diffusion models can generate very realistic images and other data.
• Mode collapse is a problem that can occur with other generative models, where the model
generates the same few samples over and over again. Diffusion models are less likely to
suffer from this problem.

Image generation models are GANs (e.g., BigGAN, ProGAN), VAEs (e.g., DRAW,
BetaVAE), and Diffusion models (e.g., DALL-E 2, Stable Diffusion, Imagen)
Image Generation Models
Art and design: Creating new forms of art and design, such as
digital paintings and fashion designs.
Entertainment: Generating special effects and animations for
movies and games.
Product development: Designing new products and prototypes.
Transformer Models
Transformer models combine an encoder-decoder architecture with a text-processing
mechanism.
An encoder converts raw, unannotated text into representations known as embeddings;
The decoder takes these embeddings together with previous outputs of the model, and
successively predicts each word in a sentence.
The encoder learns how words and sentences relate to each other, building up a powerful
representation of language without having to label parts of speech and other grammatical
features.
After these powerful representations are learned, the models can later be specialized—
with much less data—to perform a requested task.
Transformer Models

Transformers process words in a sentence simultaneously,

Enable text processing in parallel, speeding up training.
Earlier techniques like recurrent neural networks (RNNs) processed words one by one.
With transformers, one model trained on a massive amount of data can be adapted to
multiple tasks by fine-tuning it on a small amount of labeled task-specific data.
Transformers are like a supercomputer for language
Most large language models (LLMs) are based on transformer architectures.

Imagine you're trying to understand a story.

• Traditional way: You read each word one by one, trying to remember what came before.
• Transformer way: You can look at the whole story at once, seeing how the words fit
together and understanding the meaning.
• Translate languages: Turn English into Spanish, or French into Chinese.
• Summarize text: Make a long article shorter and easier to understand.
• Answer questions: Answer questions about a piece of text.
• Create new text: Write stories, poems, or even code.
General-Purpose LLMs

• GPT-4: Developed by OpenAI:

• Text generation: Creating human-quality text, such as articles, essays, and code.
• Question answering: Providing informative and comprehensive answers to questions.
• Machine translation: Translating text from one language to another.
• LaMDA: Developed by Google, focus on conversational AI. It can be used to create
chatbots and virtual assistants.
Specialized LLMs

BLOOM: A multilingual LLM trained on a massive dataset of text

from 49 languages. It is particularly effective for tasks that
require understanding and generating text in multiple languages.
Megatron-Turing NLG: A large-scale LLM developed by
Microsoft and Nvidia, designed for generating long-form text,
such as articles and reports.
Codex: A LLM specifically trained on code data, making it well-
suited for tasks like code generation and completion.
Retrieval Augmented Generation (RAGs)
combination of traditional information retrieval techniques and
generative language models
improve the quality and relevance of generated text by
incorporating factual information from external sources.
Useful for private or confidential data
Problems with Deep Learning
Computationally intensive - requiring powerful hardware such as Graphics
Processing Units (parallel processing and graphics) and Tensor Processing
Units (optimized for ML and DL training and inference)
Massive data for training.
Time consuming.
Black box - difficult to interpret, making it challenging to understand how
they arrive at their decisions.
Bias - perpetuate biases present in the training data, leading to unfair or
discriminatory outcomes.