AI & Machine Learning

Creating Intelligent Systems — from the mathematics of learning to frontier research and ethics.

New to AI? Start with the simplified version — no math required. Come back here when you're ready for the technical details.

Artificial Intelligence refers to computer systems that perform tasks typically requiring human intelligence — visual perception, speech recognition, decision-making, and natural language understanding. Its sub-fields nest: machine learning (systems that learn from data), deep learning (neural networks with many layers), and NLP (understanding human language). At its core AI is powered by mathematics; understanding it isn’t just academic — it helps you build better systems, diagnose problems, and push boundaries. This reference introduces mathematical concepts as they are needed, always starting from practical motivation.

Three Ways to Read This Material

This site covers AI at three depths. Pick the one that matches where you are:

Page Audience What it covers
AI Fundamentals (Simplified) Beginners Plain-language intuitions, no math required — the gentle on-ramp.
This Complete Reference (you are here) Practitioners The full technical treatment: learning theory, architectures, generative models, frontier research, and ethics — split across the pages below.
AI Deep Dive (Lecture) Advanced A focused lecture on transformers, LLM internals, and current research.

For practical, hands-on generative-AI guides (Stable Diffusion, ComfyUI, LoRA training), see the AI/ML section. For the complete index of every AI resource on the site, see the Artificial Intelligence hub.


How to Use This Hub

This reference is split into focused pages so you can go as deep as you need on any one topic without wading through the rest. Each page is self-contained, but they build on one another. If you are not sure where to begin, use these start here pointers:

Suggested reading order (front to back): FoundationsCore AlgorithmsDL TheoryDL ArchitecturesLoss FunctionsFine-TuningReinforcement LearningGenerative ModelsFrontier & Ethics.


Explore the Reference

Page Group What it covers
Machine Learning Foundations Foundations Statistical learning theory, optimization, kernels and SVMs, Gaussian processes, and variational inference.
Core ML Algorithms Foundations The classical workhorses — regression, trees, boosting, SVMs, k-NN, and clustering — that still win most tabular problems.
Deep Learning Theory Deep Learning Why deep networks can fit anything, how gradients flow, and what makes them generalize despite being overparameterized.
Deep Learning Architectures Deep Learning From the MLP to convolutions, recurrence, attention, Transformers, and the sequence models that followed.
Neural Network Architectures Deep Learning A single-narrative overview tying the foundations to CNNs, RNNs/LSTMs, Transformers, and multimodal models.
Loss Functions & Objectives Training Regression, classification, contrastive, ranking, and generative losses — and the why behind each.
Fine-Tuning & Transfer Learning Training Adapting pretrained models — full fine-tuning, LoRA, instruction tuning, RLHF, and DPO.
Reinforcement Learning Training Learning to act from reward — MDPs, value and policy methods, deep RL, and the algorithms behind agents and RLHF.
Generative Models Generation & Frontier Diffusion models, GANs, VAEs, and autoregressive/LLM generation.
Frontier Research & Ethics Generation & Frontier Scaling laws, mechanistic interpretability, emergent abilities, AI safety/alignment, ethics, and governance.

Types of AI

Narrow AI (weak AI) performs specific tasks within a single domain, often surpassing human performance there, but cannot generalize across domains. Nearly every deployed AI system today is narrow:

  • IBM’s Deep Blue — chess computer that defeated world champion Garry Kasparov in 1997.
  • Google’s AlphaGo — Go-playing AI that defeated world champion Lee Sedol in 2016.
  • Amazon’s Alexa / Apple’s Siri — voice-controlled virtual assistants.
  • OpenAI’s ChatGPT (GPT-4-class / o-series) — language models with multimodal input and step-by-step reasoning.
  • Claude 4 (Anthropic) — Constitutional AI with strong safety alignment and coding ability.
  • Google’s Gemini — multimodal model processing text, images, audio, and video natively.

General AI (strong AI, or artificial general intelligence/AGI) would perform any intellectual task a human can, with broad world understanding and the ability to learn and adapt across challenges. Status: not yet achieved — an active research area. The central open challenges are scalability (handling vast knowledge and reasoning), transfer learning (applying knowledge from one domain to unfamiliar ones), and commonsense reasoning (understanding everyday situations).

Machine Learning: Teaching Computers to Learn

Machine learning is a branch of artificial intelligence that focuses on the development of algorithms and models that can learn from data and make predictions or decisions. The primary goal of machine learning is to enable computers to improve their performance on a task over time without being explicitly programmed.

Types of Machine Learning

Supervised Learning

The algorithm is trained on a labeled dataset, where the input features are mapped to output labels. The goal is to learn a function that can make accurate predictions for new, unseen data.

Cat Dog Cat Model ?
Regression Classification

Unsupervised Learning

The algorithm is trained on an unlabeled dataset, and the goal is to find patterns, relationships, or structures within the data.

Discovered Patterns
Clustering Dimensionality Reduction

Reinforcement Learning

The algorithm learns by interacting with an environment, receiving feedback in the form of rewards or penalties, and adjusting its actions to maximize cumulative rewards over time.

Agent Environment Action Reward
Game Playing Robotics

AI, ML, and DL Relationship

Artificial Intelligence
Machine Learning
Deep Learning

Deep learning—machine learning with neural networks many layers deep—is where most of today’s breakthroughs happen. Deep Learning Theory explains why these models work, and Deep Learning Architectures walks through the building blocks themselves; the Machine Learning Foundations page covers the statistical learning theory and optimization that underpin both.


Key Takeaways

  • Learning is optimization. Training reduces to following the gradient downhill: $\theta_{t+1} = \theta_t - \eta\nabla_\theta\mathcal{L}$ — everything else is architecture and data.
  • Depth builds abstraction. Deep networks learn hierarchical features; the transformer’s self-attention made long-range, parallel modeling practical and now dominates language and vision.
  • Generative models reverse a known process. Diffusion models learn to denoise via $\mathcal{L} = \mathbb{E}[\lVert\varepsilon - \varepsilon_\theta(\mathbf{x}_t,t)\rVert^2]$, turning random noise into structured images.
  • Scale is predictable — to a point. Loss follows power laws in parameters and data ($L = E + A/N^\alpha + B/D^\beta$), but data quality and compute-optimal allocation matter as much as raw size.
  • Capability and responsibility scale together. Fairness, interpretability, privacy, and safety are core engineering requirements, not optional add-ons.

See Also

Reference pages in this section:

Related pages: