AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Latent Reasoning in TRMs is Secretly a Policy Improvement Operator

Arip Asadulaev, Rayan Banerjee, Fakhri Karray, Martin Takáč

ICML 2026 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add Deep Improvement Supervision (stepwise intermediate targets) to TRM, compact modelTRM-compact 0.8M · ARC-AGI-1 (pass@2)accuracy: 12.0 -> 24.0Improved
add DIS to medium TRM with same reduced recursion (T=1, n=2)TRM-medium 7M · ARC-AGI-1 (pass@2)accuracy: 27.1 -> 40.0Improved
add DIS to medium TRMTRM-medium 7M · ARC-AGI-2 (pass@2)accuracy: 0.0 -> 2.5Improved
replace TRM with DIS at full sizeTRM 7M · ARC-AGI-2 (pass@2)accuracy: 3.3 -> 6.0Improved
replace TRM with DIS at full sizeTRM 7M · ARC-AGI-1 (pass@2)accuracy: 40.4 -> 41.3Improved
reduce recursion from T=3,n=6 (336 steps) to T=1,n=2 (18 steps) with DIS, remove halting headTRM on N-Queens · N-Queens 8x8accuracy: 0.69 -> 0.69No clear effect
add DIS on compact model for ARC-AGI-2TRM-compact 0.8M · ARC-AGI-2 (pass@2)accuracy: 0.0 -> 0.0No clear effect
use integer time-step index instead of continuous diffusion time conditioning (ablation)DIS · ARC/N-QueensaccuracyImproved
DIS training with fixed 6 supervision steps, no halting headTRM training · training time on 4 H100training time: 72 -> about 40 hoursImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy