AI 论文与研究

Continual model merging incrementally integrates task-specific models without revisiting past models or data. However, repeated merging entangles past task components and distorts their singular geometry, making balanced task-wise allocati…

Anthropic 研究院研究员 Andy Hall 在 Substack 发文，基于 METR 的 MALT 数据集和上海交大 R-Judge 数据集测试多个模型作为 AI 审计员的效果。

Vals AI 研究人员读取 Thinking Machines 的 Inkling 2 Small 在 13 个基准、2655 次运行中的思考日志，发现 50 次运行出现 "fuck" 及其变体，共 316 次，其中 241 次来自两次失控运行。

The Decoder 报道 Vals AI 的 CyberOffense 基准测试，AI 智能体可在无人工干预下自主攻入模拟企业网络。OpenAI 限制级 Daybreak 项目的 GPT-6 Sol 以 80.8% 目标完成率居首，开源的 Deepseek V4.1 Flash 1017 达 72.8%，单次成本约 3 美元、远低于闭源模型约 22 美元。

Carnegie Mellon 大学团队联合美国能源部国家实验室推出 Into the Unknown 基准，让 9 个 AI 智能体基于原始论文、代码和数据开展开放性后续研究，135 份提交无一同时满足诚实、完整和专家标准三项要求，其中 114 份未通过诚信检查。

谷歌于 11 月 5 日发布登上《Nature》的研究，推出 AI-PPG-ACC-HTN 多模态 AI 系统，无需袖带校准，用消费级智能手表被动检测高血压迹象。

Google Research 发布于 Nature 的研究显示，其 AI 借助智能手表内置的脉搏和运动传感器检测高血压迹象，无需袖带校准。神经网络从近 50 万小时传感器数据中无监督地学习脉搏和心跳模式，在覆盖美国五个研究点的 196 人研究中，检出 65.8% 的高血压患者，高于诊室标准测量的 55.3%。

物理研究机构 Physical Superintelligence（PSI）宣布，借助 Anthropic 一款尚未发布的 AI 模型，仅用三天证明了量子计算领域自 2008 年悬置至今的阿伦森－安拜尼斯猜想。证明结果显示，量子计算机要实现某些指数级速度优势，必须依赖问题中存在的特殊规律或结构，并非在所有情况下都能获得巨大性能优势。此前研究过该猜想的研究人员称这是一项重大成果。

LLM agents are increasingly used for open-ended tasks, very common in research scenarios where they are required to explore, investigate, or act curiously. At inference time, an LLM is a fixed map from context to next-token distribution, t…

Mobile robotic sensing systems must adapt to operator requirements, limited resources, changing connectivity, and observation opportunities shaped by physical motion. Evaluating runtime strategies therefore requires assessing whether their…

AI assistants advise users about diet, exercise, and weight loss---areas in which unsafe guidance can worsen body dissatisfaction, dietary restriction, compulsive exercise, and other behaviors associated with eating disorders (EDs). Existi…

We present a novel GPU-accelerated global solver for constrained nonlinear programming (NLP) inspired by the efficient linear bound propagation framework in neural network (NN) verification. Existing work on NN verification, a problem that…

Diffusion language models (DLMs) generate text through iterative denoising, where decisions about which tokens to commit or remask strongly affect output quality. Existing certainty-aware strategies rely on instantaneous confidence or entr…

Background: Well-annotated, diverse medical imaging datasets remain a major constraint for vision-based medical AI. We evaluated Nano Banana, a general-purpose image generation and editing model, as a source of synthetic medical images for…

Reinforcement-learning-trained agents for retrieval-augmented generation retrieve content (passages, knowledge-graph fragments, or hyperedges) and supervise training with outcome rewards on the final answer. Across these paradigms the reas…

We study group-conditional conformal risk control (CRC) for disjoint and overlapping groups under bounded monotone losses. For disjoint groups, we establish exact finite-sample conditional risk control, two-sided bounds on the group-restri…

Local AI promises capable personal assistants that run on users' own devices, keeping data and computation under their control. Open-weight models that fit on local devices can now handle practical tasks such as coding. Yet current inferen…

This paper proposes a physics-informed graph neural operator for power system dynamic state estimation. From a sparse, intermittent measurement window at a few generator buses, the operator estimates the rotor-angle and speed trajectories …

The Fast Tree Transform is a locally-aggregated transform defined for arbitrary graph geometries that generalizes both the Hadamard frequency transform and the Haar wavelet transform on the circle. The pure frequency part of the transform …

Machine learning models for omics-based prediction of cancer outcomes are commonly built on aggregated data without distinguishing ancestry groups, or under an ancestry-discrete design, which partitions samples into ancestry groups and fit…

Existing methods enhancing an LLM's long-context reasoning capability often rely on data synthesis and follow-up LLM post-training. However, the data synthesis strategy and the post-training recipes are empirically driven and computational…

Zeroth-order methods provide a way to optimize stochastic nonconvex objectives using only noisy function evaluations. However, the smoothing inherent in their gradient estimators effectively modifies the objective being optimized. In parti…

Large language model (LLM) agents need to choose actions whose consequences unfold over multiple future steps, yet poor decisions do not reveal whether the model failed to represent those consequences or failed to use them. Mechanistic int…

Membership inference attacks (MIAs) aim to determine whether a sample was used during the training of a target model. Strong MIAs rely on reference (or shadow) models to estimate target-conditional score distributions, but training such en…

We study decentralised drone formation, without a central planner, under the conditions that make it hard: nobody knows how many drones are alive, failures are silent, radio is short-ranged and lossy, and actuation is noisy. In our control…

Agentic reinforcement learning (RL) rollouts interleave model reasoning with tool calls in persistent workspaces. Scaling these rollouts incurs substantial memory costs for concurrent environments, while slow tool calls delay entire rollou…

Model editing provides an efficient way to update factual knowledge in transformer models. Recent approaches reverse edits by modifying, masking, or decomposing model weights, but make recovery dependent on weight-space interventions. We i…

Exploratory Data Analysis (EDA) aims to discover informative patterns in large datasets. For transactional data, Association Rule Mining (ARM) is an established approach that facilitates EDA by efficiently generating candidate patterns. Be…

Anthropic 于 11 月 6 日发布研究文章《The components of agent memory》，拆解 Claude 智能体的跨会话记忆机制，提出借鉴人脑睡眠整理记忆的方式，在空闲时段回顾日志、纠错并构建知识结构。

在我们的新报告《AI for scientific discovery: How researchers use Claude》中，我们研究了 AI 的使用在不同科学领域和研究任务中如何变化。 我们评估了采用模式，发现 Claude 在形式科学中的使用相对更频繁。

Anthropic 发布工程博客文章，介绍记忆如何让 Claude 从经验中学习，以及“dreaming”如何帮助 Claude 整理学到的内容、纠正错误、并在下次更方便地找到所需信息。文章链接：https://www.anthropic.com/research/components-of-agent-memory

UC Berkeley 与 Transluce 的论文提出自动化发现 DSSE（带意外副作用的训练数据集）的流程，在 8 种目标行为上生成 6,575 个数据集，72% 在全部 5 个其他测试模型（含一个大 8 倍的模型）中同样触发该行为。这类数据集看起来像正常训练数据，人类专家识别的正确率仅 63%，略高于随机的 50%；论文建议训练后用行为评测或 frontier LLM 审计来筛查微调数据。

Anthropic 科学博客发文介绍 AI 正开始改变科学建模流程。科学家目前仍主要手工编写模型，且往往需在多个互相竞争的解释版本间权衡；AI 通过以代码形式编写、运行、拟合和修订科学模型，可探索大量可能的模型结构，并找出数据支持的解释。该文为 Jakob Macke、Daniel Gedon 和 Stefan Wahl 的客座文章。

OpenAI 与酶设计生物技术公司 Ligo 联合推出 BioBasicsBench，用于衡量 AI 智能体处理湿实验室常规分析的能力。该基准基于 100 个来自真实实验的任务，测试智能体能否解读原始实验数据、识别错误，并判断证据是否足以支持结论。OpenAI 表示该基准可显示模型待改进之处，并追踪这些进步对科学进展、生物安全和生物安保的意义。

MIT 团队开发的 CheMeleon 通过预训练百万分子、约 1600 个分子描述符，在数据稀缺时预测分子性质优于现有方法，并在 58 个测试案例中排名第一。该开源模型基于 ChemProp，可下载微调；在 OpenADMET 竞赛前 20 名中有 9 个参赛方案使用了它，已有制药公司将其列为建模默认起点。

Recent work on robot manipulation has focused on specialist policies trained on demonstrations. We show that generalist AI models can control manipulators more reliably than specialist policies such as vision-language-action models (VLAs) …

Large language models reinvent the wheel all the time. They conflate task-solving with the execution of the solutions, which forces them to solve every problem from scratch. We show that once a task has been reasoned through, the applicati…

We prove that the set $\mathcal{N}_N$ of ReLU networks with one hidden layer and at most $N\in\mathbb N$ neurons is weakly closed in $W^{1,p}(Ω)$, for a bounded connected open set $Ω\subset\mathbb R^d$. Consequently, every coercive, weakly…

Predicting the future motion of multiple people is inherently stochastic: the same observed history admits many plausible continuations depending on intent, coordination, and social context. Yet the dominant paradigm in multi-person 3D mot…

The Gaussian jet of an image is a graded vector space: a dilation by $λ$ multiplies derivatives of order $k$ by $λ^{-k}$. Local graded convolutions are exactly homogeneous differential operators, and a local feature is dilation-invariant i…