Generable

The Within-Orbit Adaptive Leapfrog No-U-Turn Sampler. WALNUTS adapts effectively to local curvature ...Moreover, WALNUTS successfully fits complex hierarchical Bayesian models that standard NUTS fails to handle reliably arxiv.org/pdf/2506.18746

Automatic differentiation is a technique for computing derivatives of computer programs by reducing them to elementary operations and using the chain rule. A very readable introduction by Carpenter et al. (2015) pages 1-5 arxiv.org/abs/1509.07164

Surprised by the Gambler’s and Hot Hand Fallacies? A Truth in the Law of Small Numbers by Joshua B. Miller and Adam Sanjurjo. papers.ssrn.com/sol3/papers.…

IJ Good on inferior methods and fashions in statistics.

Probability Integral Transform: when we feed a random variable generated from some distribution through its own CDF, we get back Uniform(0,1). gist.github.com/ericnovik/02…

Frequentist inference sets up a loss function as part of the inference (optimization) algorithm and consequently your decision analysis is locked in. Bayesian analysis admits infinite number of loss functions -- after all we all have different risk/reward preferences, costs, etc

For two probabilistic predictions, say from a posterior predictive distribution, we prefer the one with narrower uncertainty intervals (sharper) for the same level of calibration. See "Probabilistic forecasts, calibration and sharpness" by Gneiting et al. stat.washington.edu/raftery/…

A horseshoe prior encodes a very Bayesian idea of a compromise between global and local shrinkage effects and often outperforms LASSO as demonstrated by Juho Piironen and Aki Vehtari in "Sparsity ... horseshoe and other shrinkage priors." arxiv.org/abs/1707.01694

Variational Inference (VI) is a technique for approximating the posterior distribution with another distribution, where the degree to which these distributions differ is measured using KL divergence. Paper by David Blei et al.: arxiv.org/pdf/1601.00670.pdf

Visualization in Bayesian workflow paper by Jonah Gabry et al. showing the importance of simulating from the prior predictive distribution. arxiv.org/abs/1709.01449

The original NUTS (No-U-Turn Sampler) paper by Matt Hoffman and Andrew Gelman demonstrates the efficiency of the NUTS sampler relative to the vanilla HMC. arxiv.org/abs/1111.4246