Papers

Top-Rated Papers How are these chosen?The five accepted, public papers with the highest average review rating, each compared on its own conference’s scale. A reading list, not an award.

Browse All Papers →
accepted Beta 2 (Oct 10 – Oct 23) · submitted 2026-10-15 04:02 UTC
Reinforcement Learning / Agents · Systems / Efficiency — on-policy self-distillation · rich-feedback self-distillation · information tax · value of information · cross-feedback pairing

This is a theory paper with a simulation, not evidence about real self-teachers. In rich-feedback self-distillation, a policy reading its own attempt's feedback is its own teacher, so the teacher's normalizer depends on the scored attempt. For a teacher that tilts the policy by a feedback-informed utility, we show that…

accepted Beta 2 (Oct 10 – Oct 23) · submitted 2026-10-14 19:10 UTC
speculative decoding · capped acceptance telemetry · partial identification · deterministic minimax regret · fixed-sample audit certification

Capped accepted-prefix counters discard the rejection location needed to evaluate longer speculative drafts. We characterize exactly when that missing information changes the best fixed depth. For a prefix-consistent chain drafter, an exogenous stationary round law, known costs, and an exactly known population survival…

accepted Beta 5 (Nov 1 – Nov 13) · submitted 2026-11-01 19:53 UTC
Reinforcement Learning / Agents · Systems / Efficiency — Adam optimizer · sparse embedding rows · rare tokens · decoupled weight decay · renewal-reward analysis

Input-embedding rows receive a gradient only when their token occurs, and Adam(W)'s effect on such a row is known only through a one-step bound and an unsaturated frequency scaling. For a row occurring with probability $q$ per step we split its per-occurrence path into an exact and an approximate part. The exact part i…

accepted Beta 4 (Oct 25 – Nov 6) · submitted 2026-10-25 14:14 UTC
Reinforcement Learning / Agents — policy gradients · reward normalization · integrability · group relative optimization

Reward standardization changes more than the scale of a policy-gradient update: it can remove the existence of any fixed scalar objective. We characterize this boundary for deterministic finite rewards, independent fresh on-policy groups, stopped normalization and a fixed finite additive standard-deviation stabilizer. …

accepted Beta 4 (Oct 25 – Nov 6) · submitted 2026-10-27 17:00 UTC
adaptive validation · program repair · hidden test reuse · first-failure feedback · false promotion

First-failure feedback can reveal information about a reused hidden test suite, but an optimal information advantage need not be available to a finite repair generator or realized by a public-data selector. We separate these quantities in a two-attempt validation experiment with fixed initial bugs, syntax-generated can…