Relative Value Learning

July 23, 2026 ยท Grace Period ยท ๐Ÿ› ICLR 2026

โณ Grace Period
This paper is less than 90 days old. We give authors time to release their code before passing judgment.
Authors Marc Hรถftmann, Jan Robine, Stefan Harmeling arXiv ID 2607.21120 Category cs.LG: Machine Learning Cross-listed cs.AI Citations 0 Venue ICLR 2026
Abstract
In reinforcement learning, critics typically estimate absolute state values $V(s)$, estimating how good a particular situation is in isolation. However, it turns out that only differences in value are relevant for control. Motivated by this, we propose Relative Value Learning (RV), a framework that learns value differences directly via an antisymmetric function $ฮ”(s_i, s_j) = V(s_i) - V(s_j)$. We introduce a pairwise Bellman operator and prove it is a $ฮณ$-contraction with a unique fixed point equal to the true value differences, derive well-posed $1$-step, $n$-step and $ฮป$-return targets and reconstruct generalized advantage estimation from pairwise differences to obtain an unbiased policy-gradient estimator (R-GAE). Beyond theoretical results, we integrate RV with PPO and achieve competitive performance on the Atari benchmark (49 ALE games) compared to standard PPO, indicating that relative value estimation is an effective alternative to absolute critics.
Community shame:
Not yet rated
Community Contributions

Found the code? Know the venue? Think something is wrong? Let us know!

๐Ÿ“œ Similar Papers

In the same crypt โ€” Machine Learning