TY  - RPRT
TI  - Train for Many, Update with One: Improving Multi-Sample Reasoning via Fenchel-Bregman Posterior Learning
AU  - Sungjun Lim
AU  - Adrian Robert Minut
AU  - Nico Carlsen Daheim
AU  - Mohammad Emtiyaz Khan
AU  - Kyungwoo Song
AU  - Thomas Möllenhoff
PY  - 2026
UR  - https://arxiv.org/abs/2610.17735
ID  - 2610.17735
ER  -