AI 900 - 2 - The depths of machine learning

Introduction

We started our journey getting a hang of the basics related to AI, Responsible AI and Machine learning. You hear the buzz words Large Language Models (LLM's) all the time? Lets explore how they came into existance in this blog!

Type of Machine Learning

There are 2 types of machine learning based on the training data included to train the algorithm.

1. Supervised Machine Learning: In this type of machine learning the training data includes both the feature values and known label values. They are used to train models by determining a relationshop between features and labels in the past observations so that unknown labels can be predicted for features in future cases. There are 2 types of supervised machine learning: Regression (label predicted by the model is a numeric value and Classification (label predicated is a categorization or class, which inturn has binary classifcation which predicts one of the two possible outcomes and multiclass classification which predicts multiple possible values)
2. Unsupervised Machine Learning: This type of machine learning involves training models using data that consists only of feature values without any known lables. There is 1 type of unsupervised machine learning called Clustering which identifies similarities between observations based on their features and groups them into discrete clusters

Some examples of different types of machine learning below:

• Number of ice creams sold on a given day based on temperature, rainfall and wind speed = Regression (supervised machine learning)
• Patients at risk for diabetes based on blood glucose level = Binary classification (supervised machine learning)
• Identify the species of penguines based on flipper length = Multiclass classification (supervised machine learning)
• Group the flowers based on the number of leaves and petals = Clustering (unsupervised machine learning)

Deep Learning

Deep learning is an advanced form of machine learning that tries to emulate the way human brian learns. It creates an artificial neural network that stimulates electrochemical activity in biological neurons using mathematical functions. Models produced by deep learning are called deep neural networks

Transformers

Transformer models are trained with large volumes of text enabling them to represent the semantic relationships between words and use those relationships to determine probable sequences of text that makes sense.

Large Language Models are based on Transformer architecture which uses and extends some techniques that have been proven successful in modelling vocabularies to support Natural Language Processing tasks and specifically generating language.

The Transformer architecture consists of 5 components:

1. Training text of large volumes that is used to train the model
2. Encoder that breaks the text tokens which are sequenced using attention
3. Embeddings which is a collection of vectors (multi valued numeric arrays) in which each element of the vector represents a sematic attribute of the token
4. Decoder generates an appropriate output by creating a new sequence of tokens using the embeddings created by the encoder.
5. Attention technique to predict an appropriate completion of the sentence by analyzing the input tokens and semantic attrinutes embedded in the embeddings.

Conclusion

The blog post sets the baseline of the machine learning concepts and elemental architecture of how large language models are based on.

Read more about all this on Microsoft Learn - Introduction to AI in Azure

1. Module 1: Introduction to AI concepts
2. Module 2: Introduction to machine learning concepts

All the best with your learning journey, connect with me and stay tuned for the next blog in this series.

Blog 1: AI 900 - 1 - The Basics

Cheers - Kamal