Probability Theory for Artificial Intelligence logopeterindia.net
Knowledge Hub

Probability Theory for Artificial Intelligence

How probability connects to AI — the language of uncertainty behind learning, generative models, language-model sampling, decision making and trustworthy predictions.

74 topics•9 categories•Updated October 2026

Why Probability Theory Matters for AI

3 entries

Probability is the mathematics of uncertainty, and machine learning is full of it. Training by maximum likelihood or maximum a posteriori estimation is probabilistic inference; the usual loss functions (squared error, cross-entropy) come from assumed noise models. Bayesian methods quantify what a model does not know; generative models such as VAEs and diffusion models learn probability distributions; language models output a distribution over the next token; reinforcement learning is built on Markov decision processes; and calibration and conformal prediction ask whether a stated confidence can be trusted. This page collects free books, courses, papers and tools for each of these areas.

01Probability theoryOverview of the mathematical foundations: sample spaces, random variables and distributions. 02Bayes' theoremUpdating beliefs with evidence, the core of Bayesian machine learning. 03Maximum likelihood estimationChoosing model parameters that make the observed data most probable.

Foundations & Learning Resources

13 entries

Core ideas and where to learn them.

04Probability and Information Theory (Deep Learning book, ch. 3)Concise probability chapter written for machine learning readers. 05Mathematics for Machine LearningFree book with a chapter on probability and distributions. 06Probabilistic Machine Learning: An Introduction (Murphy)Free textbook taking a probabilistic view of machine learning. 07Probabilistic Machine Learning: Advanced Topics (Murphy)Companion volume on inference, generative models and decision making. 08Pattern Recognition and Machine Learning (Bishop)Classic Bayesian-flavoured textbook, free to download from Microsoft Research. 09Information Theory, Inference, and Learning Algorithms (MacKay)Free book linking probability, information theory and learning. 10Statistics 110: Probability (Harvard)Introductory probability course with lectures and problem sets. 11Introduction to Probability (MIT OCW 6.041)Free university course on probabilistic systems. 12Seeing TheoryVisual introduction to probability and statistics. 13Think Bayes (Allen Downey)Free book teaching Bayesian statistics with Python. 14Probability and Statistics (Khan Academy)Gentle introduction to probability and distributions. 15Central limit theoremWhy sums of random variables tend towards a normal distribution. 16Law of large numbersWhy sample averages converge, the basis of Monte Carlo methods.

Bayesian Inference & Probabilistic Models

8 entries

Representing beliefs, reasoning under uncertainty and fitting models with Bayesian methods.

17Probabilistic Graphical Models (Koller & Friedman)Reference text on Bayesian networks and Markov random fields. 18Bayesian networkDirected graphical models of conditional dependence. 19Gaussian Processes for Machine Learning (Rasmussen & Williams)Free book on Gaussian processes. 20Markov chain Monte CarloSampling from complex posterior distributions. 21Variational Inference: A Review for StatisticiansOverview of approximating posteriors by optimization. 22Bayesian Data Analysis (Gelman et al.)Standard Bayesian statistics text with a free PDF. 23Statistical Rethinking (McElreath)Course and book on Bayesian modelling. 24Bayesian Methods for HackersNotebook-based introduction using PyMC.

Probabilistic & Generative Deep Learning

8 entries

Neural networks that model distributions.

25Auto-Encoding Variational BayesThe variational autoencoder paper. 26Denoising Diffusion Probabilistic ModelsDiffusion models as learned reverse of a noising process. 27Score-Based Generative Modeling through SDEsUnifies diffusion and score-based models. 28Normalizing Flows for Probabilistic Modeling and InferenceReview of invertible models with exact likelihoods. 29Generative Adversarial NetworksOriginal GAN paper. 30Weight Uncertainty in Neural NetworksBayes by Backprop for Bayesian neural networks. 31An Introduction to Variational AutoencodersTutorial by the VAE authors. 32Understanding Diffusion Models: A Unified PerspectiveTutorial deriving diffusion models from VAEs.

Uncertainty, Calibration & Conformal Prediction

8 entries

Knowing how confident a model should be.

33On Calibration of Modern Neural NetworksShows deep networks are often overconfident and how to fix it. 34Dropout as a Bayesian ApproximationMonte Carlo dropout for uncertainty estimates. 35Deep EnsemblesSimple and scalable predictive uncertainty estimation. 36What Uncertainties Do We Need in Bayesian Deep Learning?Aleatoric versus epistemic uncertainty. 37A Gentle Introduction to Conformal PredictionDistribution-free prediction sets with coverage guarantees. 38MAPIEPython library for conformal prediction. 39Uncertainty ToolboxMetrics and visualizations for predictive uncertainty. 40Calibration (statistics)Background on calibrated probability forecasts.

Probability in Language & Sequence Models

6 entries

How token distributions are modelled and sampled.

41Attention Is All You NeedTransformer paper whose decoder outputs a distribution over tokens. 42The Curious Case of Neural Text DegenerationIntroduces nucleus (top-p) sampling. 43Language Models are Few-Shot LearnersGPT-3 paper on autoregressive language modelling. 44PerplexityStandard probabilistic measure of language-model quality. 45Softmax functionTurning scores into a probability distribution. 46Hidden Markov modelClassic probabilistic model for sequences.

Decisions, Bandits & Filtering

6 entries

Probability for sequential decision making and state estimation.

47Reinforcement Learning: An Introduction (Sutton & Barto)Free textbook built on Markov decision processes. 48A Tutorial on Thompson SamplingBayesian approach to the exploration-exploitation trade-off. 49Bandit Algorithms (Lattimore & Szepesvari)Free book on bandit theory. 50Markov decision processFramework for decisions under uncertainty. 51Kalman filterRecursive state estimation with Gaussian noise. 52Particle filterSequential Monte Carlo state estimation.

Libraries & Probabilistic Programming Tools

10 entries

Software for probability, sampling and Bayesian modelling.

53SciPy statisticsDistributions, tests and sampling in Python. 54PyMCPython library for Bayesian modelling and MCMC. 55StanProbabilistic programming language with Hamiltonian Monte Carlo. 56NumPyroProbabilistic programming on JAX. 57PyroProbabilistic programming on PyTorch. 58TensorFlow ProbabilityProbabilistic modelling and inference in TensorFlow. 59ArviZDiagnostics and visualization for Bayesian models. 60Turing.jlProbabilistic programming in Julia. 61GPyTorchScalable Gaussian processes in PyTorch. 62scikit-learn naive BayesSimple probabilistic classifiers.

Related Pages

12 entries

Neighbouring pages on this site.

63MathematicsMathematics resources. 64Linear AlgebraLinear algebra resources. 65Number Theory for AINumber theory for AI. 66Machine LearningMachine learning hub. 67Deep LearningDeep learning hub. 68Generative AIGenerative AI hub. 69Diffusion ModelsDiffusion models. 70Causal AICausal AI. 71Reinforcement LearningReinforcement learning resources. 72Time Series Forecasting ModelsProbabilistic forecasting models. 73Python Libraries for XAIExplainability libraries. 74Artificial IntelligenceAI hub.

Frequently Asked Questions

Why is probability important for AI?

Probability provides the language for uncertainty. Learning from noisy data, estimating parameters, generating new data, predicting the next token, choosing actions and reporting confidence all rely on probability distributions and Bayes' theorem.

What probability topics should I learn first for machine learning?

Random variables and common distributions, expectation and variance, conditional probability and Bayes' theorem, maximum likelihood estimation, and the law of large numbers and central limit theorem. Then move on to Bayesian inference, sampling methods and information theory.

What is the difference between frequentist and Bayesian approaches?

Frequentist methods treat parameters as fixed and probability as long-run frequency, while Bayesian methods treat parameters as random variables with a prior that is updated to a posterior using data. Many machine learning methods can be read either way, for example regularization as a prior.

How do generative models use probability?

VAEs, normalizing flows, diffusion models and autoregressive language models all learn a distribution over data, then sample from it. They differ in whether the likelihood is exact, approximated or implicit.

What is the difference between aleatoric and epistemic uncertainty?

Aleatoric uncertainty is noise inherent in the data and cannot be reduced by more data, while epistemic uncertainty comes from limited knowledge of the model and can shrink with more data. Methods such as ensembles, Bayesian neural networks and conformal prediction address different parts of this.

How current is this page?

Links were checked in October 2026. This is an introductory map, not a textbook: it points to free books, courses, papers and libraries. Sites that block automated checks are kept and may need a manual click.