Machine learning

Notes on Machine Learning

This collection of links serves as a comprehensive guide to various aspects of machine learning and artificial intelligence, covering everything from foundational concepts to advanced techniques and frameworks.

Machine Learning Overview

• The difference between AI and Machine Learning: Clarifying misconceptions and highlighting the distinctions between AI and ML.
  • The difference between AI and Machine Learning

Core Concepts

• Data Handling and Generation: Techniques for synthetic data creation are crucial for training models where real data is scarce.
  • Synthetic Data Generation: reference/site/Synthetic data with SDV and Gaussian copulas, reference/site/Synthetic data with SDV and CTGAN, reference/site/Synthetic data with SDV and CopulaGAN
• Model Explainability: Making machine learning models understandable to humans.
  • reference/site/Explainability
• Handling Drift: Maintaining model performance over time by addressing concept, model, and data drift.
  • Introduction to Concept Drift in Machine Learning, Model drift, Data drift

Model Development and Evaluation

• Optimisation Techniques: Algorithms to improve model performance, including both gradient-based and gradient-free options.
  • Gradient Descent, Stochastic Gradient Descent, Stochastic Gradient descent with momentum, Mini-Batch Gradient Descent, Adagrad, RMSProp, AdaDelta, Adam, reference/site/Gradient-free optimisation
• Model Selection and Evaluation: Strategies for selecting the best model and assessing its performance.
  • Cross-validation, Kernel functions: Interpretation and applications, Kernel functions
• Error and Performance Metrics: Metrics to evaluate model errors and performance.
  • Error metrics: Distance metrics
  • Performance: Language performance metrics, Model performance metrics

Machine Learning Methods and Techniques

• Time-Series Analysis: Analysing and predicting data that changes over time, and detecting anomalies in streaming data.
  • reference/site/Time-series analysis, Streaming anomaly detection
• Clustering: Grouping data points based on their similarities.
  • reference/site/K-means clustering
• Fairness: Ensuring models do not perpetuate biases.
  • Fairness in Machine Learning, reference/site/Model fairness
• Transformations: Preprocessing steps for effective algorithm performance.
  • reference/site/Feature scaling
• Recurrent Neural Networks (RNN): Recognising patterns in sequences of data with LSTM networks.
  • LSTM

Machine Learning Applications and Frameworks

• Supervised and Unsupervised Learning: Differentiating these foundational approaches with examples.
  • Supervised methods: Random Forest, Regression: reference/site/Gaussian Process Regression
  • Unsupervised methods: Self-organising maps

LLMs (Large Language Models)

• Overview of LLMs: Understanding the architecture and capabilities of large language models, including their applications in natural language processing and beyond.
  • Introduction to LLMs, Applications of LLMs
  • Retrieval-Augmented Generation (RAG): Enhancing LLMs with external knowledge retrieval
• LLM Evaluation and Benchmarking: Methods for assessing the performance of LLMs, including metrics and benchmarks relevant to their capabilities.
  • Tying into existing evaluation metrics: Language performance metrics, Model performance metrics
  • LLM Evaluation: LLM evaluation
• LLM Security and Safety: Exploring vulnerabilities, attack vectors, and defence mechanisms for large language models, including techniques to prevent misuse and ensure responsible deployment.
  • Adversarial Attacks: Understanding how malicious inputs can manipulate model behaviour through techniques like prompt injection and jail-breaking.
    • Universal and Transferable Adversarial Attacks, Prompt injection attacks
  • Detection and Prevention: Methods for identifying and mitigating security threats to LLMs in production environments.
    • Task drift detection, LLM security monitoring
  • Alignment and Safety: Approaches to ensure LLMs behave according to human values and resist harmful outputs.
    • RLHF for alignment, Constitutional AI, LLM safety benchmarks
• Machine Learning Frameworks and Tools: Tools and frameworks for ML development, from model building to deployment.
  • KServe, Cookiecutter Data Science, reference/site/Scikit-learn, Model serving

Fundamental Theory and Statistics

• Statistics in Machine Learning: Statistical concepts and methods crucial for ML algorithms and evaluation.
  • Statistics, reference/site/Streaming statistics, Thompson sampling, Statistical dependence