These are my working notes on various topics in machine learning and AI, from foundational ideas and recent research. I write them for myself as a thinking tool - they are opinionated, sometimes incomplete, and updated over time. They are also heavily interlinked, so following connections is often the best way to explore.

• Adaptive Learning Rate Optimizers
• Autoencoders
• Autoregressive Models
• BERT
• BLEU
• BM25
• Backpropagation
• Boltzmann Machines
• CNNs for NLP
• Calibration
• Capsule Networks (CapsNet)
• Challenges of GAN
• Compositional semantics and sentence representations
• Compressed Sensing
• Conditional GAN
• Confident Learning - Principled Data Cleaning
• Contrastive Divergence
• Coreference Resolution
• Counterfactual Evaluation and LTR
• Cross Validation
• Cross entropy
• Decision Theory
• Depth and Trainability
• Discrete Fourier Transform
• Discriminant Functions
• Disentangled Representations
• Distant Supervision
• Dropout
• Dyna-Q - Planning and Learning
• Dynamic Programming (RL)
• Emergent Misalignment in LLMs
• Energy based models
• Ensemble Methods
• Equivalent Kernel
• Expected Reciprocal Rank
• Fisher Information
• GRU
• Gaussian Mixture Model
• Generative Adversarial Networks
• Graph Convolutional Networks (GCN)
• Group Equivariant Convolutional Neural Networks
• Harris Corner Detection
• Hopfield Networks
• Hough Transform
• Incremental Implementation of Estimating Action Values
• InfoGAN
• Inverse Reinforcement Learning
• Jensen's Inequality
• Jensen–Shannon Divergence
• K-Means
• KL Divergence
• Kernel Methods
• LSTM
• Lagrange Multipliers
• LambdaRank
• Latenent Variable Models
• Learning to Defer
• Learning to Rank
• Least squares for classification
• ListNet and ListMLE
• Logistic Regression
• Loss Functions
• MAML - Model-Agnostic Meta-Learning
• Maximum Entropy Principle
• Maximum Likelihood Estimation
• Maximum Mean Discrepancy (MMD)
• Meta Learning
• Model Based Reinforcement Learning
• Model Complexity and Occams Razor
• Model Free Reinforcement Learning
• Monte-Carlo Tree Search
• Multi-Armed Bandits
• Multi-Head Latent Attention (MLA)
• Natural Policy Gradient
• Normalization
• Normalizing Flows
• Off-policy learning with approximation
• On-policy learning with approximation
• Online Evaluation and LTR
• PGT Actor-Critic
• Perceptron
• PixelRNN
• Policy Gradient
• Polyloss
• Positional Encoding
• Principle Component Analysis (PCA)
• Prioritized Sweeping
• Probabilistic Generative Models
• RankNet
• Recurrent Neural Networks (RNN)
• Regularized Least Squares
• Reinforcement Learning Problem Setup
• Semi-Markov Decision Processes
• SentencePiece - Unigram LM Encoding
• Similarity Measures
• State Update Functions in Partially Observable MDP
• Stochastic Gradient Descent
• Uncertainty in Machine Learning
• Variational Autoencoders
• Variational Inference
• Weight Initialization in Deep Neural Networks
• Why Generative Models
• Why implicit density models
• PPO - Proximal Policy Optimization
• TRPO - Trust-Region Policy Optimization
• Hierarchical Reasoning Model (HRM)
• MAP-Elites
• Challenges of optimizing deep models
• Tiny Reasoning Model (TRM)
• Deep Supervision with Recursion
• Backpropagation Through Time (BPTT)
• Temporal Difference Learning
• Transformers
• Direct Preference Optimization (DPO)
• Self-Attention Mechanism
• Autoregressive Generation and KV Caching in Transformers
• Beam Search
• Class Imbalance
• Grouped Query Attention (GQA)
• Mixture of Experts in Transformers (MoE)
• Mixture of Experts
• REINFORCE - Score Function Estimator
• Rotary Position Embeddings (RoPE)
• Byte Pair Encoding
• Focal Loss
• Tokenization
• RLHF - Reinforcement Learning with Human Feedback
• Bradley-Terry Model
• Importance Sampling
• Distribution Shift
• RMSNorm
• Maximum A Posteriori (MAP)
• Conformal Prediction
• Convolutional Neural Networks (CNN)
• Layer Normalization
• Monte-Carlo Estimation
• Gaussian Processes
• Stochastic Gradients
• Bias vs Variance in Machine Learning
• Convolution
• Pathwise Gradient Estimator
• Reinforcement Learning
• Advantage Functions
• Bellman Equation and Value Functions
• Control Variates
• Deep Q-Learning
• Eligibility Trace
• Expectation Maximization
• Markov Decision Processes
• Markov Reward Processes
• Multi-Network Training with Moving Average Target
• Partial Observability
• REINFORCE - Monte Carlo Policy Gradient
• Singular Value Decomposition (SVD)
• Support Vector Machines (SVM)
• Bayesian Estimation
• Collaborative filtering
• High-Dimensional Dot Product Normalization
• Activation Functions
• Bayesian Linear Regression
• Bayesian Model Selection with Model Evidence
• ReLU
• Basis Functions
• Generalized Advantage Estimate
• Monte-Carlo RL Methods
• Gaussian Distribution
• Covariate Shift