AI 论文与研究

上海理工大学顾敏院士、张启明教授团队在《自然·光子学》发表单光束多维光存储技术，单个三维存储点可区分1024种状态，对应10 bit信息容量，DVD尺寸介质理论容量约0.4 Pb，读取可靠性超过99.99%。

Hugging Face 发布论文 ProvenanceGuard，一个面向 MCP 智能体的生成后验证层，可在不重训智能体的前提下读取 MCP 调用轨迹，逐条核查答案中的声明是否由其所声称的来源支持，以捕捉"跨来源混淆"错误。

清华 NLP（OpenBMB 成员）联合中科院大学、东北大学、UIUC 和约翰霍普金斯大学提出 One-Shot OPD，将 OPD 训练集压缩到一条查询：数学任务上从 59.1 提升至 68.5（300 步），达到全量数据 OPD 69.8 的 87% 增益，并在代码、指令跟随和智能体工具使用上跨 Qwen、Llama、OLMo 成立。

小红书 AllSpark 团队提出 PACT（Policy Aligned Critic Training），通过三个正则条件唯一确定长程 Agent 的信用分配，并用 BCE 回归有界 Value、采用 Actor-Then-Critic 更新顺序改进 Critic 与策略对齐。

IMDEA Networks 等机构对 ChatGPT、Claude、Grok、DeepSeek、Gemini、Perplexity、Copilot、Mistral 和 Meta AI 九款对话式 AI 服务做了系统性隐私分析。

Unsupervised tabular anomaly detection (TAD) aims to identify anomalous rows in tabular data using normal training samples. While conventional methods rely on dataset-specific training and configuration search, recent tabular foundation mo…

Covariate effects vary across contexts and shift over time, requiring forecasters to assess how to use them for each forecasting context. As forecasting proceeds, observations for earlier forecasts become available, providing feedback on p…

System One models such as Jev offer an efficient alternative to generative language models for tasks that require decisions rather than open-ended responses. However, existing Jev models exhibit limited Chinese-language decision accuracy, …

Repeated verifier calls are useful only when they contribute conditional information. We introduce VStress, an auditable replay contract, and VStress-CA, a correlation-aware allocation policy that estimates the conditional marginal informa…

Manipulating next-token probabilities during generation can bypass the safety alignment of large language models. Existing approaches, however, rely on access to model weights or numerical token probabilities and therefore do not apply to …

Distributed inference depends on GPU collective communication that must keep pace with evolving hardware and specialized workloads. However, existing collective implementations often couple semantics, orchestration (where and when data mov…

Production RL for language models lets the sampler fall behind the learner and repairs the resulting mismatch with a truncated importance weight. We ask how long the sampler can go without a refresh under that correction, and find a cliff:…

Large language models (LLMs) excel at token-level generation but may learn undesirable abstract semantics and lack comprehensive perception. LLM-JEPA mitigates this by aligning different views of the same underlying knowledge via a joint-e…

Simulation-based inference is challenging when many heterogeneous observations must be composed, hierarchical latent structure must be preserved, and the simulator is misspecified relative to observed data. We develop sampling and fine-tun…

We study the learning dynamics of fine-tuning a policy model on self-generated and reward-weighted data, with particular focus on a generalized version of REINFORCE -- referred to as RE(S) -- that updates the rollout distribution once ever…

Group-relative policy optimization relies on reward-derived advantages and sequence-level likelihood weights, both of which can be sensitive to localized outliers. Extreme rewards can collapse the contrast among clean responses after group…

Learning neural-network models of dynamical systems with safety guarantees is a fundamental requirement for their deployment in safety-critical settings. Safety is commonly established by proving the invariance of a desired subset in state…

GUI agents automate tasks on digital devices by grounding language instructions in visual interfaces. Existing group-relative reinforcement learning improves GUI action prediction by comparing the rewards of multiple responses sampled from…

Human image animation aims to transfer motion from a driving video to subjects in a reference image. Despite remarkable progress in video generation, achieving high-fidelity animation of multiple interacting subjects remains a challenge. M…

Long-context reasoning is essential for complex and long-horizon tasks, yet the performance of large language models (LLMs) degrades as context length increases. Recent approaches address this by processing input chunk by chunk while maint…

Traffic signal control (TSC) is essential for improving urban mobility and reducing congestion. Although roadside cameras are widely deployed at signalized intersections and provide rich visual observations of evolving traffic, existing TS…

Recently, Group Relative Policy Optimization (GRPO) and its variants have been developed for policy optimization and demonstrated notable performance gains. However, these methods usually incur substantial computational overhead due to per…

Reproducibility is essential for scientific research, yet prior work shows that LLM outputs vary with hardware and batching. We identify an overlooked factor: the hidden injection of the current date into system prompts, which users cannot…

Many computer tasks recur: the same workflow runs many times, with new inputs and from different starting states. Current computer-use agents re-plan every step of every run, which makes them costly and unreliable on such tasks. We introdu…

Recent power measurements provide valuable information for photovoltaic(PV) power forecasting, but directly extrapolating short-term trends can introduce substantial errors over longer forecast horizons. To address this challenge, we propo…

You Only Edit Once： 通过局部示例精修激发 LLM 的上下文能力 挑选最佳少样本示例是一种缓慢的 System-2 搜索：组合爆炸，且往往需要反复调用 LLM。 我们把它变成了 System 1。⚡ Jev-LDE，一个 1.7B 的编辑器，扫一眼检索到的示例，只做一次编辑。LLM 只回答一次。 平均 1-shot 准确率 81.2 → 88.1 You Only Edit Once 🧵 动画演示（示意示例）。查询：“How far is it from…

苏黎世联邦理工学院软体机器人实验室基于常规仿人机械手，用强化学习训练出可依靠指尖自行行走的机械手，整机重 818 克，搭载树莓派机载计算机可独立运行。该机械手在 NVIDIA Isaac Lab 仿真平台完成训练，已在沥青、草地、碎石等 14 种表面爬行，25 次被碰倒中 21 次能自主站起；手指仍可恢复普通操作，敲键盘正确率达 90%，并通关推箱子游戏、定位精度达毫米级。

MIT 研究者 Brian Hedden 与 Manish Raghavan 系统评估了针对算法单一栽培的主要反对意见，发表于 Philosophical Perspectives。

Amazon AGI 提出 AutoGym，可从少量领域种子或过往模型轨迹出发，同时生成任务、可执行环境和验证器，用于为智能体自动构建 RL 训练环境。作者将其定位为"blueprint-first"的可验证 agent gym 生成方法，并强调创建 RL 环境正成为一项重要的 AI 工程技能。

据彭博社报道，20 多名 AI 行业领袖在一篇新论文中警告，用 AI 自动化下一代模型研发可能导致技术进步突然加速的智能爆炸，其速度或超过社会适应能力。联署者包括 Anthropic 联合创始人杰克·克拉克、OpenAI 首席科学家雅库布·帕乔茨基、杰弗里·辛顿、约书亚·本吉奥等。

研究者提出 Hindsight-Divergence Localization（HDL），利用后见诱导的 token 对数似然变化来选取分支点，只从关键位置生成续写并仅用新生成的后缀更新策略。在数学、代码和智能体任务上，相比同等组规模与训练步数的 GRPO，HDL 生成 token 最多减少 2.5 倍、rollout 墙钟时间提速 1.8 倍，智能体任务性能最高提升 12.5 分。

OTRetarget 提出一种统一方法，可从人类演示中联合重定向机器人与多物体运动，通过带符号距离、最近表面点和相对方向表示表面交互，并用熵最优传输在人体、机器人与物体几何间迁移这些量。

研究者提出长时视频掩码 Transformer（LVMT），用轻量 GRU 时序传播模块自适应选择跨时间保留的对象信息，并引入 Truncated Query Propagation（TQP）训练策略，按帧块处理视频、仅在块内反向传播，从而在长视频上实现更长时序监督且不爆显存。

研究者提出长视频记忆框架 Grounded Entity Biographies（GEB），将跨片段的同一物理实例的视觉观测归组为可检索的"传记"，并在问答时与情节证据一同检索，让模型沿身份链接追踪实体。

FocusVTC 通过自适应分辨率打破视觉文本压缩的固定分辨率权衡，在保留通用多模态能力的同时压缩输入长度。在 RULER v1 上以 72 DPI 取得 87.4 分、2.9 倍输入压缩，超过 Glyph 的 57.5 分；LongBench 得分 56.40 高于文本输入基线的 55.86，MRCR 宏平均提升 13.91 分，在线端到端延迟较纯文本提速 2.79 倍。

SkillGym 提出一套自动流水线，从互联网抓取技能并筛选可离线复现的工作流，通过 builder-reviewer 流程构建四类难度可控任务，每个任务配有参考解和可执行验证器，共建成 6.8k 个环境、收集 19k 条验证成功的轨迹用于监督微调。

研究者推出 RLE-Bench，一个覆盖交互控制、策略学习、感知与估计、机械设计四类机器人开发流程的基准，用于评测编程智能体的工程能力。该基准按任务特定指标评估智能体提交的产物，并汇总为 RLE Index，同时给出各流程的能力画像，还通过案例研究分析智能体的行为与局限。

研究者提出 ActFirst-OPD，一种"先行动、后推理"的 on-policy 蒸馏训练框架，将环境交互与完整回复生成解耦：学生模型借助参考条件逆动力学推断并执行动作，当实际转移偏离参考轨迹时切换为自主预测下一动作，再异步生成完整 think-then-act 回复接受 token 级教师监督。

WorldLine 是一个动作驱动的视觉模拟器，将可迁移的动力学学习与异构动作对齐解耦，基于超过 10,000 小时无动作机器人视频学习操作动力学，并用超过 2,000 小时、覆盖十余种本体的动作轨迹完成对齐。

研究者提出 RouteFM，通过跨异构路由环境的片段式预训练学习可复用的 LLM 路由能力，让冻结的路由器仅靠上下文即可适配新环境，而非绑定固定模型身份。在未参与预训练的 MMR-Bench 上，每个候选模型仅用 8 条观测，RouteFM 就比最强基线高出 2.23 个质量分。代码已公开。