AI and Machine Learning overview

91 topics in 14 lessons, each in one line. Every topic opens the page that teaches it.

After this course you can

  • Frame a business problem as a machine learning task, prepare its data, and judge the model with the right metric.
  • Explain how a neural network and a large language model work, from tokens and attention to temperature.
  • Design an AI application with prompts, RAG, tools and agents, choose between prompting, RAG and fine-tuning, and run it safely.

1What AI and machine learning are

2A first model: linear regression

3Classification

4Data and generalisation

5Trees, forests and clustering

6Neural networks

7From model to production

8How large language models work

9Prompting

10Retrieval-augmented generation

11Agents and tools

12Customising and running models

13Responsible AI and security

14A support assistant over company documentsWorked case

Sources and credit

43 sources

The chapters and drawings are written by NeuroqAI from these sources. Each lesson lists its own at the end of its chapter, and figures on the pages are taken from them or computed from them.

Amazon Web Services8 sourcesAWS Certified AI Practitioner (AIF-C01) exam guideThe domains this course covers for AWS AILessons 1, 3, 4, 8, 9AWS Certified Machine Learning Engineer Associate (MLA-C01) exam guideThe tasks of the ML engineer examLessons 2, 3, 4, 7, 8Built-in algorithms (Amazon SageMaker AI)Which ready-made algorithm fits which problemLessons 2, 3, 5What is Amazon SageMaker AI?Training, deploying and monitoring your own modelsLessons 4, 7, 12What is Amazon Bedrock?Hosted models, customisation and throughput optionsLessons 7, 12Amazon Bedrock Knowledge BasesManaged retrieval over your own documentsLesson 10Responsible AIAWS's dimensions of responsible AILesson 13Fairness, model explainability and bias detection with SageMaker ClarifyBias metrics and SHAP explanationsLesson 13
Standards, research and other publishers35 sourcesExam AI-901 study guide · MicrosoftThe skills measured for Microsoft's AI fundamentals examLesson 1Mastering the game of Go with deep neural networks and tree search (2016) · Silver et al., NatureAlphaGo: learning beat hand-written rulesLesson 1Training language models to follow instructions with human feedback (2022) · Ouyang et al.How a model is taught to follow instructionsLessons 1, 8Deep Learning · Goodfellow, Bengio and CourvilleThe free textbook: training, regularisation, networksLessons 2, 4, 6Random Search for Hyper-Parameter Optimization (2012) · Bergstra and Bengio, JMLRWhy random search beats a gridLessons 2, 4The Elements of Statistical Learning, 2nd edition · Hastie, Tibshirani and FriedmanThe free standard text on the classic modelsLessons 2, 3, 5Random Forests (2001) · Leo Breiman, Machine LearningMany trees on random samples, votingLessons 2, 3, 5XGBoost: A Scalable Tree Boosting System (2016) · Chen and GuestrinThe boosted trees that win on tablesLessons 2, 3, 5An introduction to ROC analysis (2006) · Tom Fawcett, Pattern Recognition LettersROC curves and AUC, clearly explainedLessons 3, 4, 8ROUGE: A Package for Automatic Evaluation of Summaries (2004) · Chin-Yew Lin, ACLScoring a summary against a referenceLessons 3, 4, 8BLEU: a Method for Automatic Evaluation of Machine Translation (2002) · Papineni et al., ACLScoring a translation against a referenceLessons 3, 4, 8BERTScore: Evaluating Text Generation with BERT (2020) · Zhang et al.Scoring by meaning rather than matching wordsLessons 3, 4, 8Parquet file format · Apache ParquetRow groups, column chunks and statisticsLessons 4, 7Learning representations by back-propagating errors (1986) · Rumelhart, Hinton and Williams, NatureThe method every network still trains withLesson 6NVIDIA H100 Tensor Core GPU · NVIDIAThe chip most large models are trained onLesson 6LoRA: Low-Rank Adaptation of Large Language Models (2021) · Hu et al.Fine-tuning by training a small add-onLessons 7, 12QLoRA: Efficient Finetuning of Quantized LLMs (2023) · Dettmers et al.Fine-tuning a large model on one GPULessons 7, 12Attention Is All You Need (2017) · Vaswani et al.The transformer, behind every large language modelLesson 8Language Models are Few-Shot Learners (2020) · Brown et al.GPT-3: 175 billion parameters, and learning from examples in the promptLesson 8The Llama 3 Herd of Models (2024) · MetaHow an open model is trained on 15 trillion tokensLesson 8Efficient Estimation of Word Representations in Vector Space (2013) · Mikolov et al.Words as vectors: where embeddings beganLesson 8Denoising Diffusion Probabilistic Models (2020) · Ho, Jain and AbbeelHow image models make pictures from noiseLesson 8Prompt engineering overview · AnthropicSystem prompts, examples, XML tags and letting the model thinkLesson 9Structured outputs · OpenAIAnswers that always match a JSON schemaLesson 9OWASP Top 10 for Large Language Model Applications · OWASPThe main risks when software uses an AI modelLessons 9, 13, 14Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (2020) · Lewis et al.The paper that named RAGLessons 10, 14Efficient and robust approximate nearest neighbor search using HNSW graphs (2016) · Malkov and YashuninThe index most vector databases useLesson 10Reciprocal Rank Fusion outperforms Condorcet and individual rank learning methods (2009) · Cormack, Clarke and Buettcher, SIGIRMerging keyword and vector results with k = 60Lessons 10, 14Model Context Protocol · Model Context ProtocolHow models reach tools and data: hosts, clients, serversLesson 11Agent2Agent (A2A) protocol · Linux FoundationHow agents from different makers work togetherLesson 11Building effective agents · AnthropicWorkflows versus agents, and five patterns that workLesson 11ReAct: Synergizing Reasoning and Acting in Language Models (2022) · Yao et al.Think, act, observe: the loop most agents runLesson 11Responsible AI principles · MicrosoftFairness, reliability, privacy, inclusiveness, transparency, accountabilityLesson 13AI Risk Management Framework · NISTGovern, map, measure, manage for AI systemsLessons 13, 14Path to high-quality LLM-based Dasher support automation · DoorDash EngineeringA RAG assistant with a two-tier guardrail and an LLM judgeLesson 14