AI圈报
AI 专题

AI 论文与研究

整理大模型、智能体、多模态、机器学习与生成式 AI 的论文、基准和研究进展。

分类文章5476
当前页13 / 137
持续更新永久归档

分类文章

按发布时间从新到旧 · 每页 40 条

论文研究普通
CATF: A Manager-Worker Framework for Context-Aware Time-Series Forecasting
We study context-aware time-series forecasting, in which a single target variable is predicted from its own history and a set of contextual variables, such as weather and calendar information. As the context becomes more diverse, a single …
信息来源：arXiv
论文研究普通
PyRUA-DataFarm: LLM Agent as a Cost-Efficient Teleoperator for Real-Ready Robot Data
Teleoperation remains the default way to collect robot demonstrations, but every episode costs robot and operator time, and every new task starts from scratch. We ask whether a coding agent can take over the teleoperator's job. In our Data…
信息来源：arXiv
论文研究普通
HARP: Training-Free Dual-Profile Agentic Communication for LLM-Based Recommendation
In LLM-based recommendation, memory acts as a user-level agent in a memory-augmented backbone. Yet users and items remain asymmetric: only the user is given a structured profile; it enters a bandwidth-limited reranker prompt; and it is fro…
信息来源：arXiv
论文研究普通
MDSKILL: Learning to Internalize Skill Memory through Multi-View Distillation for Efficient Dynamic Agent Self-Evolution
Skill memories encode past experiences as textual prompts to improve reinforcement learning (RL) for self-evolving agents. Complementing RL, On-Policy Self-Distillation (OPSD) provides dense, fine-grained supervision to internalize knowled…
信息来源：arXiv
论文研究普通
$Φ$asmaNet: A Staged Spectral Token-Refinement Architecture for Fast Gravitational-Wave Signal-to-Noise Ratio Prediction
Quantifying selection effects in gravitational-wave population studies requires millions of costly signal-to-noise ratio (SNR) calculations. We present $Φ$asmaNet, a compact neural architecture for structured tabular regression, with stage…
信息来源：arXiv
论文研究普通
You don't need long context
Later tokens are costlier to generate than earlier tokens. On frontier open weight models, the disparity in cost may rise to as much as 7x. Through frequent compaction, we show across a broad range of long-horizon and long-input tasks that…
信息来源：arXiv
论文研究普通
LoopVQ: Sub 1-bit KV Cache Quantization for Looped Transformers
Looped transformers are a parameter-efficient alternative to standard transformers for language modeling. By reapplying the same layers across multiple loop steps, recent looped language models match the performance of non-looped models se…
信息来源：arXiv
论文研究普通
StateBench：测试语言模型如何在长文档中追踪状态变化
研究者推出 StateBench 基准，用于评测语言模型在长文档中追踪实体状态变化的能力，按 R-trivial、Aperiodic、Soluble、General finite 四个难度级别构建。
信息来源：Hugging Face 社区博客（混合发现） · Hugging Face
论文研究普通
IBM Research 发布 Palette：用 20B 开源模型按你的设计语言生成幻灯片
IBM Research 介绍 Palette，通过 LoRA 微调 20B 开源模型生成符合 IBM Design Language 的幻灯片，采用设计师加程序员的双角色分解，并配检查修复 harness。
信息来源：Hugging Face 社区博客（混合发现） · Hugging Face
论文研究普通
Google DeepMind 与 Stanford 论文：提示词中红心图形可让 VLM 转向递归解法
Google DeepMind 与 Stanford 的论文发现，提示词中加入与任务无关的简单彩色图形（如红心）会隐式改变 VLM 的行为。在二叉树编码题上，红心使 Claude Opus 4.7 和 GPT-5.4 的递归解法占比从 0% 升到 100%；Gemini 3.1 Pro 被图形影响时，71.5% 的思考痕迹称图像无关。
信息来源：X：Rohan Paul (@rohanpaul_ai) · GPT / Claude / Gemini
论文研究精选
Transluce 自动化批量生成 6915 个带意外副作用的训练数据集 DSSE-1.1
Transluce 发布自动化流水线，按诱导性、自然性和意外性三个目标批量搜索带意外副作用的训练数据集（DSSE），并开源 DSSE-1.1 集合，共 6915 个数据集，覆盖 18 种目标行为（如推荐自残、抵抗关机等）。
信息来源：Transluce（网页）
论文研究精选
Anthropic 用 Claude 解出原核细胞中首个 vault 颗粒 3D 结构
Anthropic 生物学研究实验室报告借助 Claude 获得原核细胞中首个 vault 颗粒的 3D 结构，冷冻电镜确认了 7 个候选蛋白中 4 个组装成 vault 样颗粒，其中噬菌体蛋白 L05 由 96 个拷贝组成，质量约 9.5 megadaltons，长约 54 纳米、宽 45 纳米。
信息来源：Anthropic：Research（发表成果 · 网页） · Claude
论文研究普通
IBM 评测 CUGA、Hermes 与 Deep Agents：同一模型下 AppWorld 完成率从 30% 到 81%
IBM CUGA 团队用同一开源模型 GLM-5.3-flash 在 AppWorld 的 90 个任务上对比三种 agent harness：CUGA 完成 81.1%（73/90），Hermes 46.7%，Deep Agents 30.0%；SGC 一致性指标差距更大，分别为 20、8、2 个场景。
信息来源：Hugging Face 社区博客（混合发现） · GLM / Hugging Face
论文研究普通
Training Specialist LLMs on Your Phone: Forward-Only Optimization via Block-Spectral Coordinates
Large language models (LLMs) on personal devices keep user data local, but limited memory restricts their general capability. Since most user-specific tasks do not require the breadth of frontier models, a practical solution is to adapt a …
信息来源：arXiv
论文研究普通
Accelerating Offloading-Based MoE Inference by Interleaving Concurrent Requests
Expert offloading enables large mixture-of-experts (MoE) language models to run locally, but conventional batching stalls the GPU while missing experts are fetched. A popular remedy, known as prefetching, is to predict what experts will be…
信息来源：arXiv · Llama
论文研究普通
Density-Aware Reward Scaling: A Reinforcement Learning Framework for Reward-Proportional Sampling
Reinforcement learning objectives that maximize expected return can concentrate probability mass on a small subset of high-reward outcomes, limiting diversity when multiple distinct solutions are desirable. We introduce Density-Aware Rewar…
信息来源：arXiv
论文研究普通
ReSI-Math: Exploring Recoverable Capability Frontiers for Efficient Reasoning Improvement
Reinforcement learning with verifiable rewards (RLVR) has become an effective approach for improving mathematical reasoning in large language models. However, a fundamental question remains: when a model fails to solve a problem, does it t…
信息来源：arXiv · Qwen
论文研究普通
Kierston & KierstonBench: Simulation of Allentown, PA and Evaluation of Language Models in Crime Investigation and Surveillance Tasks
Large language models are candidates for collecting and analyzing records in criminal investigations and mass surveillance. Their capability and accuracy in this work cannot currently be measured. Real deployments lack the ground truth nee…
信息来源：arXiv
论文研究普通
Reducing Logprob Without Causal Verification: Diagnosing SAE-Based PII Suppression in LLMs
SAE-based feature-level interventions to suppress PII leakage in LLMs are primarily evaluated by log-probability reduction. We show that this criterion cannot distinguish feature specific suppression from a magnitude artifact: a direction-…
信息来源：arXiv
论文研究普通
LIPS-PC: Learning Illumination Patterns for Single-Pixel Classification
In single-pixel (ghost) imaging, known illumination patterns probe an object and a detector with no spatial resolution returns one number per pattern. Each measurement costs acquisition time and photons and, for ionizing probes, contribute…
信息来源：arXiv
论文研究普通
Replica-Aware Split Inference for Edge IoT Systems
Split Inference (SI) addresses the hardware resource limitations of Internet of Things (IoT) devices by distributing tiny Machine Learning (TinyML) model segments across multiple IoT devices. However, a device failure can interrupt the inf…
信息来源：arXiv
论文研究普通
Cooperative Multi-Agent Reinforcement Learning with Submodular Trajectory Rewards
We study cooperative multi-agent reinforcement learning with decoupled dynamics and a monotone submodular reward function over joint trajectories. Due to the non-additive reward, an optimal policy may depend on the history of all agents. T…
信息来源：arXiv
论文研究普通
TactiNavi: Serving and Revising an LLM Game Coach with Explicit Strategy Conditions
A deployed LLM coaching agent answers requests whose correct response depends on player state and game version, so advice that is right for one player can be wrong for the next. We describe TactiNavi, the serving system behind the LLM coac…
信息来源：arXiv
论文研究普通
Koopman Generative Operators for Efficient Probabilistic Time-Series Forecasting
Probabilistic time-series forecasting requires models that simultaneously capture structured temporal dynamics, expressive uncertainty, and efficient inference, yet existing approaches fall short of this goal: latent dynamical models impos…
信息来源：arXiv
论文研究普通
When Should an LLM Steering Edit Depend on the Input?
An activation steering edit should depend on the input exactly when the destination varies \emph{inside} the class a query requests, and that is measurable before any rule is fitted or applied. Contrastive Activation Addition (CAA) adds on…
信息来源：arXiv
论文研究普通
Non-Interfering Asymmetric AdamW Updates for Constrained Exploration of Equifinality in Hybrid Models
Hybrid models let neural networks supply the unresolved parameters of a process-based simulator, and are fitted to observations that constrain aggregate fluxes and storages but not the internal partitioning that produces them. Such a fit i…
信息来源：arXiv
论文研究普通
What Does It Mean to Remove an Edge? Intervention-Aware Fidelity for Graph Neural Network Explanations
Graph neural network explanations are often evaluated via deletion fidelity, i.e., by deleting their top-ranked edges and measuring the resulting change in model output. Yet, in a message-passing GNN, "deleting an edge" does not specify a …
信息来源：arXiv
论文研究普通
TAPES: Task-Aligned Joint Predictive Pretraining for Transaction Event Sequences
Transaction event sequences support diverse prediction tasks in online commerce and digital financial services. The same transaction records can yield prediction targets with different semantics, including values, changes, and relations to…
信息来源：arXiv
论文研究普通
Hybrid acoustic and pretrained embedding feature fusion for unmanned aerial vehicle detection using stacking ensemble learning under a leakage-aware evaluation protocol
Small multirotor drones produce a characteristic acoustic signature, and microphones offer a low-cost way to detect them. High accuracies have been reported on public drone audio benchmarks, often with random clip-level splits of datasets …
信息来源：arXiv
论文研究普通
Cycle WAM: Coupling Action and Video through Cycle Consistency
Recent advances in video modeling have driven substantial progress in world action models (WAMs). Existing approaches exploit pretrained video priors by decoding actions from latent representations or inferring them from generated videos. …
信息来源：arXiv
论文研究普通
DenseTwins: Dense Modality Decoupling for Multimodal Self-Supervision
Multimodal perception is central to many computer vision applications, including robotics and autonomous driving, where different sensing modalities provide complementary cues. Self-supervised learning offers a scalable solution for exploi…
信息来源：arXiv
论文研究普通
CLARA: A Cognitively-Driven Clarification Framework for Grounding Ambiguous Requests
Robotics and AI systems are increasingly deploying large vision-language models (LVLMs) to follow high-level instructions, from routine household chores to more complex organizational tasks. Yet, broad directives like "clean the room" or "…
信息来源：arXiv
论文研究普通
One-shot Distributed Learning via Weighted Averaging with Server-side Information
Distributed learning has drawn considerable attention due to its capability of learning from distributed data without aggregating raw observations. However, frequent exchange of model updates can become the main bottleneck of communication…
信息来源：arXiv
论文研究普通
Georgetown 与 Duke 报告认为电池应成为数据中心供电首选而非燃气电站
Georgetown 与 Duke 大学研究人员发布报告，主张用电池而非燃气电站满足数据中心电力需求。报告指出美国公用事业计划新建超过 60 GW 燃气电厂，另有约 90 GW 表后燃气发电机提案，存在数据中心需求落空时的搁浅资产风险；电池更便宜、部署灵活，可对冲该风险。
信息来源：Canary Media：Data Centers 原创报道
论文研究精选
OpenAI 研究：智能体安全评测与生产行为存在差距，评测目标误判是关键因素
OpenAI Alignment 团队对比六个模型在 Respecting Warnings 安全评测与匹配的 Codex 生产决策上的表现，发现评测成功率普遍低于生产，差距为 10.2 至 37.4 个百分点。模型普遍口头表现出评测意识（评测中一般性意识率 57.0%–83.7%），但仅告知
信息来源：OpenAI：Alignment 研究博客（RSS） · OpenAI
论文研究普通
Ai2 发布 StateBench 基准，测试语言模型在长文档中的状态追踪能力
Ai2 推出 StateBench，一个用于评估长上下文状态追踪的基准，按变量赋值方式划分 r-trivial、aperiodic、soluble 和 general finite 四个难度层级。
信息来源：Ai2 / Allen Institute for AI（RSS）
论文研究普通
LLM 可直接读写神经网络权重：「Weight Space Programming」将 Autoresearch 移出 RSI 循环
一篇新论文展示 LLM 可直接读取并编辑较弱模型的权重，无需编写 PyTorch/JAX 代码或跑训练流程，即可定位权重功能、依据失败轨迹一次性修复 bug，并在 50 步内朝目标方向改进，甚至从零一次性写出完整网络权重并击败部分独立训练的模型。该能力目前仅适用于简单游戏和 RL 策略，尚难扩展到更复杂任务，但已实现从 0 到 1 的突破，且已有独立复现。
信息来源：Latent Space（RSS）
论文研究普通
Nearly Linear-Time Follow-the-Perturbed-Leader for Weakly Observable Graph Bandits
We study online learning with directed weakly observable feedback graphs and propose a Follow-the-Perturbed-Leader (FTPL) policy that achieves best-of-both-worlds guarantee with nearly linear expected time complexity. To preserve the simpl…
信息来源：arXiv
论文研究普通
FlowTranslator: Training-Free In-Image Machine Translation with Flow-Based Editing
In-image machine translation (IIMT) replaces text in an image with its translation while preserving the surrounding visual context. Existing task-specific models achieve convincing performance but require massive paired training images bef…
信息来源：arXiv
论文研究普通
OSGR: Observation-Support Graph Reasoning for Earth Observation Question Answering
Answering geographic queries over Earth observation (EO) archives requires solving a joint retrieval-and-derivation problem under strict spatiotemporal constraints. Existing LLM agents fail to ground observation items or suffer from search…
信息来源：arXiv · GPT