AI 论文与研究

Molecular learning models are strongly shaped by their underlying representations. Yet standard sequential and graph formalisms struggle to explicitly encode higher-order topology, such as ring systems and recurring motifs. Existing higher…

Looped Transformers achieve parameter efficiency by repeatedly executing a shared block across recurrent loops. Each loop yields an intermediate representation decodable for the same next token, yet standard decoding discards earlier state…

Quasi-Newton (QN) methods have long been among the most effective methods for large-scale unconstrained convex optimization. Two obstacles have limited their use in deep learning: non-convexity and enormous parameter sizes. We introduce So…

Current controllable video generation systems often rely on 2D motion trajectories or sparse drag signals for object motion. These controls are ambiguous because the same 2D trajectory can correspond to different 3D motions, especially whe…

Multi-teacher on-policy distillation (MOPD) aims to combine the strengths of RL-trained teachers in a single student, but how teacher signals affect parameter changes remains underexplored. We study Qwen3-1.7B with four domain teachers tra…

NVIDIA 新论文提出，与其给 AI 智能体更多选项，不如给它一个更好的裁判：让小型智能体每步生成 8 个候选命令，再由裁判模型挑选执行。用前沿强模型当裁判时，成功率从 50% 提升到 68%，且无需重新训练；若由小模型自评，提升则小得多。

营销公司 Graphite 对前沿模型写作习惯的研究发现，共 13000 个短语在 AI 内容中出现频率至少是人类内容的两倍。

卡内基梅隆、MIT、纽约大学和斯坦福的研究团队开发的不完全信息博弈 AI Ataraxos，以 15 胜 1 负 4 平击败了被认为是最强 Stratego 玩家的 Pim Niemeijer。

Meta 等机构提出 Context Language Models（CLM），把上下文作为文件让模型用 Bash 自由编辑，自主决定保留、重写或删除内容，并区分于 RLM 不允许模型直接编辑实时交互上下文的做法。zero-shot 下在 BrowseComp-Plus 上比现有上下文管理策略高 11.4% 准确率、省 21.5% FLOPs；在线 RL 使 Qwen3.5-9B 提升 47.6% 且少用 12% FLOPs；配套 Suffix Cache Reuse 相比…

Meta 等机构发布论文提出 Context Language Models（CLM），将上下文作为可读写的文件让模型用 Bash 自主编辑，决定保留、重写或删除内容。

研究人员开发的 AI 系统 Ataraxos 以 85% 的有效胜率（平局计半胜）击败了曾获 4 次世界冠军的史上最强 Stratego 选手 Niemeijer，并在 2025 世界锦标赛表演赛中取得 40 局 38 胜。

Adobe 提出把标准答案写成每次运行都重新取数的代码，再由 AI 评分器对照检查智能体回答。在 53 个测试用例上，这种做法的 AI 评分与人类专家的一致率比文字描述答案高 29%，且少用 16% token；若没有可对照的答案，AI 评分器表现还不如随机。论文题为《Skill-based Agentic Evaluation for Real-time Data Science Tasks》。

Google DeepMind 联合牛津、芝加哥大学论文探讨 AI 大量取代工作后政府应如何保障民众，结论是没有任何单一政策在所有情景下都有效。再培训等最受欢迎的思路只在 AI 持续创造新岗位时有用，否则只是重新分配剩余岗位。作者建议现在扩大失业救济与低薪税收抵免，失业持续则把抵免转为收入下限，劳动者经济份额持续下降则让公民持有公共投资基金份额。

We introduce Gacha Decoding, an inference-time method for eliciting diverse language model generations that scales with model capability. Across open-ended domains (in-the-wild chat, creative writing, planning for image generation, and pro…

Peptide sequence governs both the structures formed through supramolecular assembly and the dynamics by which they emerge, but predicting either requires resolving slow collective processes among many interacting molecules. Molecular dynam…

Attributing model behavior to synthetic training data requires knowing what produced each training item before estimating what that item caused. A waveform-label pair does not preserve this knowledge. We propose a generation-provenance sub…

We study causal logistic bandits with counterfactual fairness constraints. The causal structure is given through known factual and counterfactual feature maps that share an unknown logistic reward parameter, but the learner observes only f…

A user's movie, news, and dialogue histories differ in their native actions and outputs, yet each interaction supplies evidence that can update user memory. We study whether these histories can train one reusable update mechanism. An actio…

World models allow agents to plan in latent space by choosing a sequence of actions that most reduces the distance to a given goal state. Thus, planning can benefit from latent representations whose distances mirror commute-times in the en…

Understanding a nonlinear dynamical system from time series requires not only reproducing its trajectories, but also identifying a simple representation that preserves its essential dynamical structure. Almost-linear recurrent neural netwo…

Factories are shifting toward smaller lot sizes with high product customization, requiring frequent re-programming of flexible and reconfigurable automation systems. LLM-based agents can be deployed in two complementary roles: Offline, the…

Multilayer networks allow the modeling of interactions between the same entities across different contexts, such as temporal observations, varying settings, or interactions of different types. The goal of community detection in multilayer …

Single-particle cryo-electron microscopy (cryo-EM) has become a widely adopted technique for biomolecular structure determination. The conventional cryo-EM computational pipeline first combines many particle images to reconstruct an electr…

Stable port-Hamiltonian neural networks certify asymptotic stability by construction. Yet, their Hamiltonian is a global Lyapunov function with a single global minimum, so they can represent only dynamic systems with {one} attractor. We de…

We introduce a new framework for one-step generative modelling on finite state spaces. To extend drifting beyond continuous domains, we use discrete Wasserstein geometry to define a target-relative KL gradient flow over the transitions of …

Large language models are moving from helping write up research to helping do it, which makes it important to know how the scientific text they produce differs from human writing. Work on this question has stayed mostly at the surface, usi…

Tool-using large language model (LLM) agents turn generated text into real side effects, so poisoned tool metadata, retrieved pages, memory, and reusable skills can steer the next call. Vetting an artifact before admission does not settle …

When developers change one component of an agent, such as its controller, a learned model or its verifier, they usually judge the change by an aggregate task score. That score cannot tell whether improvement was attainable, which component…

Regulatory compliance checking - deciding whether a target document satisfies the obligations of a regulation - requires interpreting dense legal text, identifying which provisions apply, and grounding each decision in explicit evidence. W…

Reinforcement learning offers the prospect of a reusable sequential decision-making mechanism for spacecraft trajectory design, motivating policy interfaces that connect learned decisions to the underlying maneuver geometry. This paper dev…

We study the classical single-machine scheduling problem of minimizing the sum of completion times of jobs in a non-clairvoyant setting, where the processing time of each job remains unknown until its completion. This is a hard problem for…

This paper presents the European Crisis Management Ontology (ECMO), a modular OWL-based ontology intended as a cross-sectoral reference for disaster risk reduction and response. ECMO is designed to be organised as a network of ontological …

Reinforcement learning for trading often struggles to balance upside participation with drawdown control. Profit-only policies can collapse toward passive long exposure on upward-drifting assets, while aggressively risk-penalized rewards c…

Meta 论文提出 Meta-Reasoning Agent，让 worker 执行任务、独立 controller 决定下一步，在最大预算下于全部 12 项对比测试中胜过无 manager 的同款智能体。用 GPT-5.5 跑代码基准时，预算增至 3 倍，得分从 64.1% 升至 71.5%，而对照智能体停滞在 64% 附近。

ActiveSaddler 将 Agent harness 优化中"训练场景如何选择"这一维度建模为非平稳 bandit 问题，把反复出现的失败抽象为可复用的失败模式臂，动态估计各模式的潜在学习收益，在复现已知弱点与探索新场景之间自适应平衡。

研究者提出 SciUtopia，一个持续运行的闭环 LLM 智能体模拟框架，用于研究学术研究生态系统，覆盖研究方向选择、合作、投稿、同行评审、重投、引用、资助与研究者流失等环节。

AgSpec 是一个面向编码智能体流水线的检索式投机解码框架，从会话、工作区和全局语料库中检索草稿 token，并按智能体离线画像设定草稿长度上限、再依据验证反馈在线调整。在两个仓库级多智能体编码基准上，AgSpec 在多数设置下优于五种检索式草稿器和 EAGLE-3，batch size 1 时生成吞吐较自回归解码最高提升 4.37 倍，batch size 16 时最高 4.76 倍。

研究以 Qwen3-1.7B 和四个同源 RL 教师模型为对象，剖析多教师同策略蒸馏（MOPD）中教师信号如何影响参数变化，并用 SmolLM3-3B 做补充诊断。

研究者提出推理时框架 PoS，为 LLM 智能体构建并持续维护显式信念状态作为决策上下文，每个信念结合当前世界状态估计与未解决的任务需求。PoS 通过一致性校验和任务进度监控检测“Belief Trapping”，并按陷阱模式与未满足需求类型定制恢复策略。在覆盖执行与诊断的四项基准上，PoS 在三种 LLM 主干下均取得最高整体表现，消融实验验证了一致性校验与恢复机制的必要性。

OctLLM 将几何以八叉树占用 token 的显式 3D 序列输入模型，并通过随机清空倒数第二层节点、省略其后代，得到更短的 S-Octree 用于位置感知掩码建模生成与 3D 理解。