Transformers can do Bayesian Clustering

Prajit Bhaskaran & Tom Viering · 2025

Read Transformers can do Bayesian Clustering by Prajit Bhaskaran & Tom Viering (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-24318.txt.

Description

Prajit Bhaskaran and Tom Viering present Cluster-PFN, a transformer model that performs Bayesian clustering and handles missing data with proper uncertainty. Bayesian clustering accounts for uncertainty but is costly at scale, and simple filling of missing values ignores the doubt those values carry. Cluster-PFN extends prior data fitted networks to this unsupervised task.

Trained only on synthetic datasets drawn from a finite Gaussian mixture prior, the model learns to estimate the posterior over both the number of clusters and each point's assignment. It estimates cluster counts more accurately than classic criteria like AIC and BIC and variational inference, while matching or beating their clustering quality, including on data with missing entries.

Catalog record

CREATORPrajit Bhaskaran & Tom Viering
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE49 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-24318

Browse the full library · About the Tower of Babel