Why DPO is a Misspecified Estimator and How to Fix It

ICLROral2026

Authors: Aditya Gopalan, Sayak Ray Chowdhury, Debangshu Banerjee
Affiliation: Indian Institute of Science
Venue: ICLR 2026
Track: Oral

TL;DR

DPO is not sound by design and can fail due to misspecification, we fix it with careful analysis.

Opening excerpt from the authors’ abstract. source

Read the paper

OpenReview

← All ICLR 2026 Oral papers · Browse the whole archive