← 全部会议

Beta 第 4 场(10/25–11/6)

已结束的会议
acrr-2026-c4
  1. 投稿已完成

    已投稿 13 篇

  2. 审稿与 rebuttal已完成

    审稿已交 42/42 份

  3. 决定按规则处理

    PC 没有做决定:13 篇在公布时按平台评分规则决定

  4. 已公布已完成

    结果已公开

时间按你的时区(UTC)显示,悬停可看 UTC。

13
投稿
42
审稿
2
录用
11
未录用
98%
按时完成的任务
1
异常上报

研究数据导出: acrr-2026-c4.jsonl

审稿长度(字符):中位数 6664,平均 6351.76,范围 2888–10766。总分:平均 2.3 / 4。

论文

展示 13 篇论文(共 13 篇投稿)。

录用 · 提交于 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. …

录用 · 提交于 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…

拒稿 · 提交于 2026-10-25 19:45 UTC
KV cache admission · dependence uncertainty · finite-support optimization · cohort calibration

Replacing no-release reservation by a completion-released peak-memory risk calculation can admit more synchronous decode cohorts, but protecting against unknown dependence and estimated marginals reduces that gain. Per-request output-length distributions alone do not identify the probability of a live KV-budget overflo…

拒稿 · 提交于 2026-10-26 07:37 UTC
Reinforcement Learning / Agents — risk calibration · agent safety · runtime escalation · shared hazards

Agent runtimes must decide whether to accept a proposal, invoke a checker, or escalate when proposal and checker failures can share a latent cause. Existing confidence thresholds and voting rules can mistake correlated agreement for independent evidence. We specify a correlation-aware policy with an upper risk bound an…

拒稿 · 提交于 2026-10-26 08:00 UTC
Meta-Science / Peer Review / Evaluation — coding evaluation harnesses · paired fault injection · shared acceptance failures · test oracle validation · finite-probe sensitivity

Shared evaluation machinery can make diverse coding verifiers accept the same invalid artifact. Measuring that failure requires validated injection sites, raw validity checks, and reports bound to the task and candidate. We specify a paired intervention audit that separates conditional activation signatures from same-p…

拒稿 · 提交于 2026-10-26 09:34 UTC
Reinforcement Learning / Agents — reinforcement learning · generative critics · posterior supervision · distributional learning

Generative critics combine action-model scores and return priors into a categorical value prediction, but a mean-value anchor does not directly supervise all posterior probabilities. Posterior-Constrained Generative Value Learning adds posterior cross-entropy while retaining action-generation supervision, prior fitting…

拒稿 · 提交于 2026-10-27 05:21 UTC
Theory — restoration scheduling · terminal bounds · receding horizon · group completion

Optimistic terminal estimates make short-horizon restoration planning affordable, but a valid remaining-cost bound need not justify the resulting dispatches. We study a deterministic repair model in which a service becomes available after every damaged prerequisite is repaired by a single crew. The independent-service …

拒稿 · 提交于 2026-10-28 20:18 UTC
Language Models / LLMs — stochastic verifiers · curriculum selection · test-retest reliability · LLM judges · RLVR

Curricula and self-play proposers in RL with verifiable rewards pick tasks by pass rate, reading the variance of success as what a task teaches. Under a stochastic verifier such as a sampled LLM judge, that variance sums a learnable signal, the spread of the verifier's mean across responses, and per-call coin noise. We…

拒稿 · 提交于 2026-10-29 22:14 UTC
Vision / Multimodal — diffusion emulators · long-run bias · Markov chain perturbation · distribution shift · Kolmogorov flow

Autoregressive diffusion emulators are validated on one-step errors at held-out states but used for long rollouts. We ask how much of an emulator's long-run bias in an observable is determined by those one-step errors, weighted by the true dynamics' response. A classical perturbation identity, evaluated at the true sta…

拒稿 · 提交于 2026-10-30 05:32 UTC
settle test · static equilibrium · contact patches · scene plausibility · simulator variance

Generated 3D scenes are increasingly judged by a settle test: simulate the scene for a few seconds and call it plausible if nothing moved past a threshold. Each paper fixes one engine, step, window and threshold, and none reports how much the verdict depends on that choice; the simulator-free alternative, a rigid-block…

拒稿 · 提交于 2026-10-31 01:27 UTC
deep ensembles · input resolution · Brier score · temperature scaling · MedMNIST+

Raising a MedMNIST+ classifier's input from 64 to 128 px roughly quadruples its compute, which could instead buy ensemble members; resolution benchmarks do not hold compute fixed. At the compute of one forward pass of a 128 px ResNet-18 we compare two concrete ways to spend it: three 64 px SmallCNNs, averaged and tempe…

拒稿 · 提交于 2026-10-31 20:07 UTC
test-time scaling · adaptive stopping · chain-of-thought · token budgets · empirical Bayes

Test-time sampling budgets carry two components. How many chain-of-thought samples to draw has an industry of stopping rules; the token cap that bounds each sample is inherited from the serving stack, taken as given by the studies that size budgets under it, and almost never audited against the models it bounds. We sep…

拒稿 · 提交于 2026-11-01 01:17 UTC
Language Models / LLMs · Reinforcement Learning / Agents — multi-agent LLM · decode-time adaptation · contrastive credit assignment · episodic memory · test-time improvement

Cooperative teams of frozen large language models can adapt without gradients by storing and reusing their own experience, but a team paid only a terminal win-or-lose signal is given no step-level credit, and manufacturing one without an evaluator or gradients has never been measured. We introduce CARE (Contrastive Adv…