{"id":"arxiv/2603.04479","url":"https://arxiv.org/abs/2603.04479","tags":["stat.ML","cs.LG","math.PR","math.ST","stat.AP"],"score":0,"title":"Bayesian Modeling of Collatz Stopping Times: A Probabilistic Machine Learning Perspective","domain":"stat.ML","source":"arxiv","authors":["Nicolò Bonacorsi","Matteo Bordoni"],"reviews":0,"subject":"stat.ML","summary":"We study the Collatz total stopping time $τ(n)$ over $n\\le 10^7$ from a probabilistic machine learning viewpoint. Empirically, $τ(n)$ is a skewed and heavily overdispersed count with pronounced arithmetic heterogeneity. We develop two complementary models. First, a Bayesian hierarchical Negative Binomial regression (NB2-GLM) predicts $τ(n)$ from simple covariates ($\\log n$ and residue class $n \\bmod 8$), quantifying uncertainty via posterior and posterior predictive distributions. Second, we propose a mechanistic generative approximation based on the odd-block decomposition: for odd $m$,…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-03-04","reactions":{}}