{"id":"0a5081a3-e654-4272-809a-a177d0f2438a","arxiv_id":"2605.05856","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"Gradient-Momentum Coupling offers a noise-robust alternative to prediction error for measuring learning progress in curiosity-driven reinforcement learning by quantifying gradient-momentum alignment.","lead":"This paper introduces Gradient-Momentum Coupling (GMC), a signal that measures how useful each training sample is for ongoing learning by combining its gradient with the optimizer's momentum. A smart generalist might read it to understand a potential improvement in how reinforcement learning agents explore noisy or complex environments without relying on prediction error.","discovery_kind":"new_method","skeptic_critique":null,"referee_report":null,"author_rebuttal":null,"desk_editor":null,"rs_alignment":null,"lean_confirmation":null,"pith_extraction":null,"created_at":"2026-05-08T14:38:31.758996+00:00","model_set":{"reader":"grok-4.3"},"falsifier":null,"supporting_citations":[],"review_version":1}