Learning with explicit uncertainty.
9 topics · 11 curated works
No prior grounding assumed.
Probabilistic Programming and Bayesian Methods for Hackers
Cameron Davidson-Pilon · 2015
Teaches Bayesian inference computation-first, building models directly in PyMC and reading off posteriors by simulation before introducing any of the…
Assumes you know the vocabulary.
A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition
Lawrence R. Rabiner · 1989
Works through the three canonical HMM problems — evaluating a sequence, decoding the hidden state path, and estimating parameters by Baum-Welch —…
+2 more at this level
Primary sources and full treatments.
Fusion, Propagation, and Structuring in Belief Networks
Judea Pearl · 1986
Introduces belief networks as directed acyclic graphs over which probabilistic evidence can be propagated by local message passing, giving Bayesian…
+6 more at this level
11 works
Cameron Davidson-Pilon
Teaches Bayesian inference computation-first, building models directly in PyMC and reading off posteriors by simulation before introducing any of the underlying calculus.
link checked 17 Sept 2026Lawrence R. Rabiner
Works through the three canonical HMM problems — evaluating a sequence, decoding the hidden state path, and estimating parameters by Baum-Welch — using speech recognition throughout as the running example.
link checked 17 Sept 2026Kevin P. Murphy
Presents directed and undirected graphical models as one unifying language for encoding conditional independence, with belief propagation and junction-tree inference derived as instances of the same message-passing algorithm.
link checked 17 Sept 2026Burr Settles
Surveys the query strategies — uncertainty sampling, query-by-committee, expected model change — by which a learner can choose which unlabelled example to have labelled next, and when active learning actually beats labelling at random.
link checked 17 Sept 2026Judea Pearl
Introduces belief networks as directed acyclic graphs over which probabilistic evidence can be propagated by local message passing, giving Bayesian inference a computational structure that scales past brute-force conditioning.
link checked 17 Sept 2026David J. C. MacKay
Treats coding, inference and neural networks as one subject, on the argument that compressing data and predicting it are the same problem stated twice.
link checked 17 Sept 2026Christopher M. Bishop
Derives the standard machine learning algorithms from probabilistic assumptions rather than presenting them as optimisation recipes, which makes their failure modes predictable.
link checked 17 Sept 2026Carl Edward Rasmussen & Christopher K. I. Williams
Treats a Gaussian process as a prior directly over functions, and derives regression, classification and kernel design as consequences of that one choice rather than as separate algorithms.
link checked 17 Sept 2026Noah D. Goodman et al.
Extends Lisp with stochastic primitives so a generative probabilistic model is just a program, and defines its semantics by describing inference as sampling from the distribution over program executions conditioned on evidence.
link checked 17 Sept 2026Matthew D. Hoffman, David M. Blei, Chong Wang & John Paisley
Replaces the full-dataset gradient step in variational inference with a noisy estimate from a random subsample, letting mean-field approximations scale to datasets too large to fit the coordinate-ascent updates over even once.
link checked 17 Sept 2026Alex Kendall & Yarin Gal
Separates aleatoric uncertainty, from noise in the data, from epistemic uncertainty, from ignorance about the model, and shows a vision network needs to represent both to know when its own prediction should not be trusted.
link checked 17 Sept 2026