LearningTree Β· AI
Artificial Intelligence
Foundation
A structured reference covering the full spectrum of AI β from history and mathematics to deep learning, agentic systems, and responsible AI practice.
5
Tiers
12
Domains
80+
Chapters
All
Levels
This reference takes you from zero to frontier AI β structured across five progressive tiers so beginners, developers, and researchers each find their entry point and grow from there.
Tier 1 Β· Beginner β Intermediate
Foundations
History, philosophy, and mathematics β the intellectual bedrock before the algorithms begin.
Domain 01 β Foundations of AI
Where AI came from, what it actually is, the key paradigms, and the foundational frameworks that every practitioner needs before touching a single algorithm.
Ch 1.1
What Is Artificial Intelligence?
Definitions, the four capabilities (perceive, reason, learn, act), AI vs ML vs DL
Introductory Ch 1.2
History of AI β Origin to Present
Dartmouth 1956 β two AI winters β AlexNet β Transformers β ChatGPT β Agents
Core Ch 1.3
Symbolic AI & Knowledge Representation
Expert systems, logic, search algorithms, why GOFAI failed to scale
Core Ch 1.4
The Turing Test & Philosophy of Mind
Chinese Room, consciousness, symbol grounding, strong vs weak AI
In-depth Ch 1.5
Problem Solving & Rational Agents
PEAS framework, environment types, reflex β utility β learning agents
In-depth Ch 1.6
Key Paradigms & Schools of Thought
Connectionism vs symbolicism, Bayesian AI, evolutionary computation
Core Ch 1.7
The AI Landscape Today & Tomorrow
Foundation models, benchmarks, open problems: reasoning, causality, robustness
In-depth π Domain 1 β What you'll learn
- AI = optimisation toward a goal using data and computation β not magic
- Two AI winters caused by data, compute, and algorithm gaps β now all solved
- 2012 AlexNet β 2017 Transformer β 2022 ChatGPT are the three inflection points
- PEAS framework: the foundation of modern agentic AI in Domain 8
- Modern LLMs = hybrid of all paradigms β not purely one approach
Domain 02 β Mathematics & Statistics for AI
The mathematical language of machine learning β linear algebra, calculus, probability, and optimisation. You don't need to master all of it upfront, but understanding the core ideas makes every algorithm click.
Ch 2.1
Linear Algebra
Vectors, matrices, eigenvalues, SVD β the language of data
In-depth Ch 2.2
Calculus & Gradients
Derivatives, chain rule, Jacobians β how backprop works
In-depth Ch 2.3
Probability Theory
Distributions, expectation, variance, conditional probability
Core Ch 2.4
Bayesian Inference
Bayes' theorem, MLE, MAP estimation, priors
Core Ch 2.5
Information Theory
Entropy, KL-divergence, cross-entropy loss explained
Core Ch 2.6
Optimisation
Gradient descent, Adam, convexity, learning rate schedules
In-depth Ch 2.7
Graph Theory
Graphs in AI: knowledge graphs, GNNs, network analysis
Reference Tier 2 Β· Intermediate
Core Machine Learning
Traditional algorithms and the deep learning architectures that power everything in modern AI.
Domain 03 β Classical Machine Learning
Supervised, unsupervised, and ensemble methods β the core ML toolkit that still powers ~80% of production systems.
Ch 3.1
The ML Landscape
Supervised, unsupervised, reinforcement β taxonomy, bias-variance, the ML workflow
Introductory Ch 3.2
Regression
Linear & logistic regression, cost functions, gradient descent, regularisation
Core Ch 3.3
Classification
Decision boundaries, k-NN, Naive Bayes, precision/recall, ROC curves
Core Ch 3.4
Decision Trees & Ensembles
CART, Random Forests, Gradient Boosting, XGBoost β bagging vs boosting
In-depth Ch 3.5
Support Vector Machines
Hyperplanes, margins, kernel trick, soft margins, SVM vs modern methods
In-depth Ch 3.6
Clustering & Unsupervised Learning
K-Means, DBSCAN, hierarchical clustering, PCA, dimensionality reduction
Core Ch 3.7
Model Evaluation, Metrics & Validation
Cross-validation, confusion matrix, F1, AUC-ROC, overfitting, pipelines
Core π Domain 3 β What you'll learn
- Supervised learning: inputβoutput mapping via labelled data β regression & classification
- Ensemble methods (Random Forest, XGBoost) are the most-used ML in production today
- Bias-variance tradeoff: underfitting vs overfitting β the central tension in ML
- Unsupervised learning discovers structure without labels β clustering & dimensionality reduction
- Model evaluation: cross-validation + right metric matters more than the algorithm choice
Domain 04 β Deep Learning
Neural networks from perceptron to Transformer β activation functions, backprop, CNNs, RNNs, attention, and modern architectures.
Ch 4.1
Perceptron to MLP
The artificial neuron, multi-layer perceptrons, universal approximation
Introductory Ch 4.2
Activation Functions
Sigmoid, tanh, ReLU, GELU β why non-linearity matters
Core Ch 4.3
Backpropagation & Gradient Flow
Chain rule, computational graphs, vanishing/exploding gradients
In-depth Ch 4.4
Training Deep Networks
Optimisers (SGD, Adam), batch norm, dropout, learning rate schedules
In-depth Ch 4.5
Convolutional Neural Networks
Convolutions, pooling, feature maps β AlexNet to ResNet to EfficientNet
Core Ch 4.6
RNNs & LSTMs
Sequential processing, gating mechanisms, why Transformers replaced them
Core Ch 4.7
The Transformer
Self-attention, multi-head attention, positional encoding β the architecture that changed AI
In-depth Ch 4.8
Transfer Learning & Fine-Tuning
Pre-train once, fine-tune many β LoRA, adapters, the modern paradigm
Core Ch 4.9
Generative Models
VAEs, GANs, diffusion models β learning to create, not just classify
In-depth π Domain 4 β What you'll learn
- Neural networks learn via backpropagation + gradient descent β no magic
- CNNs exploit spatial locality for images; RNNs process sequences step-by-step
- The Transformer replaced both with self-attention β O(1) depth, fully parallel
- Transfer learning: pre-train on large data, fine-tune on your task β the dominant paradigm
- Generative models (GANs, diffusion) learn to create new data, not just classify existing data
Tier 3 Β· Advanced
Advanced & Specialized AI
Language, vision, and decision-making β the three major specialisations of modern AI.
Domain 05 β NLP & Large Language Models
From word embeddings to GPT-4 and beyond β tokenisation, pre-training, fine-tuning, prompting, RAG, alignment, and evaluation.
Ch 5.1
NLP Fundamentals & Text Preprocessing
Four ambiguity layers, stopwords, stemming, TF-IDF, the NLP stack
Introductory Ch 5.2
Tokenisation
BPE, WordPiece, SentencePiece β how text becomes numbers
Core Ch 5.3
Word Embeddings
Word2Vec, GloVe, vector arithmetic β king β man + woman β queen
Core Ch 5.4
Contextual Embeddings
ELMo, sentence-transformers β same word, different vector per context
In-depth Ch 5.5
The GPT Family
GPT-1 to GPT-4, scaling laws, Chinchilla, emergent abilities, open-source LLMs
In-depth Ch 5.6
BERT & Encoder Models
Bidirectional attention, MLM, RoBERTa, DistilBERT, DeBERTa
In-depth Ch 5.7
Prompt Engineering
Zero-shot, few-shot, chain-of-thought, structured prompting, pitfalls
Core Ch 5.8
Retrieval-Augmented Generation
RAG architecture, vector databases, chunking, advanced RAG patterns
Core Ch 5.9
Hallucination, Alignment & Evaluation
Why LLMs confabulate, HHH, RLHF, BLEU/ROUGE, benchmarks
Core Ch 5.10
LLM Fine-Tuning in Practice
SFT, LoRA, QLoRA, DPO β the complete practitioner's guide
In-depth π Domain 5 β What you'll learn
- BPE tokenisation: iteratively merge most frequent pairs β text β numbers
- Word2Vec geometry encodes meaning: king β man + woman β queen
- GPT = decoder-only, BERT = encoder-only β generation vs understanding
- Prompt engineering + RAG + fine-tuning = the three tools every AI practitioner uses daily
- Hallucination is a feature, not a bug β LLMs are trained for fluency, not facts
Domain 06 β Computer Vision
Image fundamentals, CNN architectures, object detection, segmentation, GANs, Vision Transformers, multimodal AI, and video/3D vision.
Ch 6.1
Image Fundamentals & Classical CV
Pixels, colour spaces, filters, edge detection, HOG, SIFT β and why deep learning won
In-depth Ch 6.2
CNN Architectures for Vision
AlexNet, VGG, Inception, ResNet, MobileNet, EfficientNet β the architecture lineage
In-depth Ch 6.3
Object Detection
R-CNN family, YOLO, SSD, anchor boxes, NMS β detecting objects at scale
Core Ch 6.4
Image Segmentation
Semantic, instance & panoptic segmentation β U-Net, Mask R-CNN, SAM
Core Ch 6.5
Generative Adversarial Networks
GAN fundamentals, DCGAN, StyleGAN, CycleGAN β learning to generate images
In-depth Ch 6.6
Vision Transformers & Modern Architectures
ViT, DeiT, Swin Transformer, ConvNeXt β attention meets vision
In-depth Ch 6.7
Multimodal AI
CLIP, DALL-E, vision-language models, GPT-4V, Gemini β images and text together
In-depth Ch 6.8
Video Understanding & 3D Vision
Optical flow, video models, depth estimation, point clouds, NeRF
Core π Domain 6 β What you'll learn
- CNNs exploit spatial locality and weight sharing β the key inductive bias for images
- ResNet solved the vanishing gradient problem with skip connections β enabled 100+ layer networks
- YOLO unified detection as a single regression problem β orders of magnitude faster than R-CNN
- Vision Transformers (ViT) prove attention works on images β treating patches as tokens
- CLIP learns visual-semantic alignment from image-text pairs β the foundation of multimodal AI
Domain 07 β Reinforcement Learning
MDPs, dynamic programming, TD learning, Q-learning, deep RL, policy gradients, PPO, and RL in real-world systems including RLHF.
Ch 7.1
RL Foundations
Agent-environment loop, MDPs, rewards, returns, policies, value functions
Introductory Ch 7.2
Dynamic Programming
Bellman equations, policy evaluation, policy iteration, value iteration
In-depth Ch 7.3
Monte Carlo & TD Learning
MC prediction & control, TD(0), TD(Ξ»), eligibility traces
Core Ch 7.4
Q-Learning & SARSA
Off-policy vs on-policy, Q-tables, function approximation, convergence
Core Ch 7.5
Deep Q-Networks
DQN, experience replay, target networks, Double DQN, Dueling DQN
In-depth Ch 7.6
Policy Gradients & Actor-Critic
REINFORCE, baseline subtraction, A2C, A3C β optimising policies directly
In-depth Ch 7.7
PPO, SAC & Model-Based RL
Proximal Policy Optimisation, Soft Actor-Critic, world models, Dyna
In-depth Ch 7.8
RL in the Real World
RLHF for LLMs, robotics, games, sparse rewards, safety in RL
Core π Domain 7 β What you'll learn
- RL = learn by doing β no labels, only reward signals from environment interaction
- Bellman equations express recursive value decomposition β the foundation of all RL algorithms
- DQN combined Q-learning with deep networks β first superhuman Atari performance
- PPO is the workhorse of modern RL β used in RLHF to align LLMs like ChatGPT
- Real-world RL: sparse rewards, safety constraints, sim-to-real gap β harder than games
Tier 4 Β· Advanced β Expert
Agentic & Systems AI
Autonomous agents that plan, act and adapt β plus the engineering discipline of deploying AI at scale.
Domain 08 β AI Agents
LLM agents with tool use, memory, planning, multi-agent collaboration, and production deployment β the frontier of AI engineering.
Ch 8.1
What Are AI Agents?
Agent definition, perception-action loop, LLM agents vs classical agents, taxonomy
Introductory Ch 8.2
Tool Use & Function Calling
APIs, code execution, web search, structured outputs, tool schemas
Core Ch 8.3
ReAct & Reasoning Loops
ReAct, Chain-of-Thought, Reflexion, self-correction, scratchpad reasoning
In-depth Ch 8.4
Memory Systems
In-context, episodic, semantic, procedural memory β long-horizon agent behaviour
Core Ch 8.5
Planning & Task Decomposition
HiGP, Tree-of-Thoughts, MCTS for agents, hierarchical task planning
In-depth Ch 8.6
Multi-Agent Systems
Agent orchestration, AutoGen, CrewAI, debate, specialisation, coordination
In-depth Ch 8.7
Evaluation, Safety & Reliability
Agent evals, prompt injection, hallucination in agents, failure modes
Core Ch 8.8
Production Agents
Observability, tracing, latency, cost, LangSmith, deployment patterns
Core π Domain 8 β What you'll learn
- An LLM agent = LLM + tools + memory + planning loop β not just a chatbot
- ReAct: interleave reasoning and acting β grounding decisions in observable tool outputs
- Memory gives agents persistence across sessions β the missing piece for long-horizon tasks
- Multi-agent systems enable parallelism, specialisation, and debate β better than one large agent
- Production agents require observability, error recovery, and cost management β not just capability
Domain 09 β MLOps & AI Engineering
Building, deploying & operating AI systems β data pipelines, experiment tracking, model serving, CI/CD for ML, monitoring, drift detection, vector databases, LLM APIs, and infrastructure.
Ch 9.1
The ML Lifecycle & MLOps Overview
ML lifecycle, MLOps vs DevOps, maturity levels, tech stack map
Core Ch 9.2
Data Pipelines & Feature Engineering
Batch vs streaming, Feast, Tecton, Great Expectations, DVC
In-depth Ch 9.3
Experiment Tracking & Model Registry
MLflow, Weights & Biases, model versioning, lifecycle stages
Core Ch 9.4
Model Serving & Deployment Patterns
Batch vs real-time, TorchServe, Triton, vLLM, canary/blue-green/shadow
In-depth Ch 9.5
CI/CD for Machine Learning
ML-specific CI/CD, Airflow, Kubeflow, Prefect, pipeline orchestration
Core Ch 9.6
Monitoring, Drift & Observability
Data drift, concept drift, Evidently, WhyLabs, retraining triggers
In-depth Ch 9.7
Vector Databases & LLM Infrastructure
Pinecone, Qdrant, pgvector, LLM gateways, LiteLLM, prompt caching
Core Ch 9.8
Containerisation, Orchestration & Cost
Docker for ML, GPU scheduling, spot instances, quantisation, cloud vs on-prem
Core π Domain 9 β What you'll learn
- 87% of models never reach production β engineering, not algorithms, is the bottleneck
- Feature stores bridge training and serving β eliminating training/serving skew
- ML CI/CD tests code + data + model quality β triggered by code, data, or drift
- Models break silently β drift detection is the early warning system
- Vector databases power RAG β HNSW index, ANN search, LLM gateway patterns
Tier 5 Β· All Levels
Applied & Responsible AI
Real-world use cases, ethical practice, and where the field is headed β the human side of AI.
Domain 10 β Ethics & Safety
Bias, fairness, explainability, privacy, AI safety, alignment, governance, disinformation, societal impact, and long-term existential risk.
Ch 10.1
AI Fairness β Bias & Discrimination
Bias sources, fairness definitions, impossibility theorem, COMPAS, mitigation strategies
Core Ch 10.2
Explainability & Interpretability
LIME, SHAP, attention maps, model cards, legal right to explanation
Core Ch 10.3
Privacy & Data Governance
LLM memorisation, differential privacy, federated learning, right to be forgotten
In-depth Ch 10.4
AI Safety β Technical Alignment
Alignment problem, adversarial robustness, specification gaming, RLHF, scalable oversight
In-depth Ch 10.5
Societal Impact
Labour markets, power concentration, environmental costs, inequality, ghost workers
Core Ch 10.6
AI Governance & Regulation
EU AI Act, US approach, international frameworks, NIST AI RMF, risk pyramid
Core Ch 10.7
Disinformation & Information Integrity
AI-generated disinformation, deepfakes, detection, C2PA provenance, elections
Core Ch 10.8
Long-Term AI Safety & Existential Risk
The expert debate, catastrophic risk scenarios, safety research, responsible development
In-depth π Domain 10 β What you'll learn
- Fairness is a value judgement β multiple definitions exist and the impossibility theorem proves they cannot all hold simultaneously
- LIME and SHAP explain individual predictions; model cards document per-subgroup performance
- Differential privacy provides provable privacy guarantees β Ξ΅ controls the privacy-utility tradeoff
- EU AI Act: risk pyramid from banned (social scoring) to minimal risk β world's first comprehensive AI law
- Long-term AI risk is genuinely contested among serious experts β not a mainstream-vs-fringe debate
Real-world AI across healthcare, finance, robotics, autonomous vehicles, education, scientific discovery, code generation, creative AI, and enterprise.
Foundation models, world models, neuromorphic computing, quantum AI, embodied AI, AGI concepts, mixture of experts, neuro-symbolic AI, edge AI, AI hardware, consciousness, and the road ahead.
Ch 12.1
Foundation Models
Scaling laws, Chinchilla, multimodal fusion, post-training, test-time compute
Core Ch 12.2
World Models & Simulation
Sora, JEPA, Genie 2, latent-space prediction, physical understanding
Core Ch 12.3
Neuromorphic Computing
Intel Loihi 2, IBM NorthPole, spiking neural networks, brain-inspired chips
In-depth Ch 12.4
Quantum AI
Qubits, NISQ era, quantum ML algorithms, quantum advantage spectrum
In-depth Ch 12.5
Embodied AI
RT-2, sim-to-real, humanoid robots, vision-language-action models
Core Ch 12.6
AGI Concepts
Definitions, DeepMind levels, paths to AGI, timeline debate, ARC benchmark
In-depth Ch 12.7
Mixture of Experts
Sparse activation, routing, Switch Transformer, Mixtral, DeepSeek-V3
Core Ch 12.8
Neuro-Symbolic AI
AlphaGeometry, LLM + code, GraphRAG, neural + symbolic reasoning
Core Ch 12.9
Edge AI & Federated Learning
On-device LLMs, federated learning, quantisation, model compression
Core Ch 12.10
AI Hardware
H100, B200, TPU v5p, Groq LPU, TSMC geopolitics, training economics
Core Ch 12.11
AI Consciousness
Hard problem, functionalism, IIT, Chinese Room, sentience debate
In-depth Ch 12.12
The Road Ahead
Open problems, reasoning, alignment, energy, governance, predictions
Core