AIF-C01-AI-ML-Concepts-Relationships.md

Comprehensive Mapping of AI, ML, Deep Learning, and Related Concepts

1. Artificial Intelligence (AI)

Definition: AI is the broad field of creating machines or systems that perform tasks requiring human intelligence, such as reasoning, learning, and language understanding.

Subfields:

• Machine Learning (ML)
• Natural Language Processing (NLP)
• Generative AI
• Computer Vision
• Robotics

2. Machine Learning (ML)

Definition: A subset of AI focused on developing algorithms that enable computers to learn from data and make decisions.

Types of Learning:

• Supervised Learning
• Unsupervised Learning
• Reinforcement Learning
• Semi-Supervised Learning
• Transfer Learning

Common Algorithms:

Supervised Learning:

• Linear Regression: Models the relationship between a dependent variable and one or more independent variables.
• Support Vector Machines (SVM): Finds the hyperplane that best separates data into classes for classification and regression.
• Convolutional Neural Networks (CNNs): Used for image and video recognition tasks.
• Recurrent Neural Networks (RNNs): Suited for sequential data, like time series or language.
• Decision Trees: A tree-like model that splits data into branches for predictions.
• Random Forest: An ensemble method combining multiple decision trees to improve accuracy.
• Neural Networks: Computational models inspired by the human brain, used in deep learning.
• Transformers: Advanced architecture for NLP tasks, e.g., BERT and GPT.

Unsupervised Learning:

• Clustering: Groups similar data points without pre-labeled categories.
• k-Means Clustering: Partitions a dataset into 'k' groups based on feature similarity.
• Principal Component Analysis (PCA): Reduces data dimensionality while preserving variability.
• Anomaly Detection: Identifies rare items or events that differ significantly from the majority of data.

Reinforcement Learning:

• Q-Learning: Learns the value of actions in a particular state, without a model of the environment.
• Deep Q-Networks (DQN): Combines Q-learning with deep neural networks.

Transfer Learning:

Knowledge gained through one task or dataset is used to improve model performance on another related task and/or different dataset.

3. Deep Learning

Definition: A subset of ML using neural networks with many layers to model complex data patterns, effective for tasks involving large datasets and complex representations.

Types of Neural Networks:

• Convolutional Neural Networks (CNNs): Primarily used for image and video recognition.
• Recurrent Neural Networks (RNNs): Well-suited for sequential data, like time series or language.
• Generative Adversarial Networks (GANs): Generates new data similar to the training data.
• Transformers: Advanced architecture used for NLP tasks like BERT and GPT.

Sampling in Deep Learning:

• Temperature: Controls the randomness in the model’s predictions. Lower values produce more deterministic outputs, while higher values increase randomness and diversity.
• Top-k Sampling: Limits the number of probable next tokens to k, introducing randomness but within a more controlled range.
• Top-p (Nucleus) Sampling: Considers tokens whose cumulative probability exceeds a certain threshold (p), offering flexibility in diversity.

4. Natural Language Processing (NLP)

Definition: A subfield of AI focused on enabling computers to read, understand, and generate human language.

NLP Tasks:

• Text Classification: Categorizes text into predefined categories.
• Language Modeling: Predicts the next word or sequence in a sentence (e.g., GPT models).
• Sentiment Analysis: Determines the sentiment expressed in text.
• Named Entity Recognition (NER): Identifies and classifies key elements in text.
• Machine Translation: Translates text between languages (e.g., evaluated using BLEU).

NLP Models:

• Traditional ML Models: Naive Bayes, Support Vector Machines.
• Deep Learning Models: RNNs, LSTMs, Transformers (e.g., BERT, GPT).

Evaluation Metrics:

• BLEU (Bilingual Evaluation Understudy): Evaluates machine-generated translations by comparing them to reference translations.
• ROUGE: Measures the overlap between machine-generated and reference summaries.

5. Generative AI

Definition: A subfield of AI focused on generating new content (text, images, audio) similar to the training data, using models like GANs and transformers.

Types:

• Text Generation: Produces new text based on input prompts (e.g., GPT models).
• Image Generation: Creates images from text descriptions or other images (e.g., DALL-E).
• Audio Generation: Synthesizes new audio, like music or speech.

Evaluation Metrics:

• BLEU: Evaluates machine-generated text quality against reference text.
• ROUGE: Assesses the overlap between machine-generated and reference summaries.

Sampling Techniques for Generative AI:

In generative models like GPT, temperature, top-k, and top-p (nucleus sampling) are critical for controlling the creativity and coherence of the generated text:

• Temperature: Controls the level of randomness in generated content.
• Top-k: Limits the next token selection to the top k most probable choices.
• Top-p (Nucleus Sampling): Selects tokens with cumulative probabilities that meet or exceed a threshold (p).

6. AWS Services Related to AI/ML:

Amazon Bedrock:

Enables access to foundation models (FMs) such as GPT, allowing integration of temperature, top-k, and top-p during text generation to fine-tune outputs.

Example Bedrock Usage:

import boto3

client = boto3.client('bedrock')

response = client.generate_text(
    model_id='gpt-model-id', # Replace with the actual model ID in AWS Bedrock
    input_text='What is the capital of France?',
    temperature=0.7,  # Balances creativity and coherence
    top_k=50,  # Consider the top 50 most likely tokens
    top_p=0.9,  # Use nucleus sampling with a threshold of 90%
    max_tokens=100  # Number of tokens to generate
)

print(response['generated_text'])