👍 72
06/04 08:00
Code language models need repository-level context to resolve imports, APIs, and project conventions. Existing methods inject this knowledge as long inputs (retrieved through RAG or dependency analysis) or through per-repository fine-tuning and LoRA -- costly at repository scale and brittle to evolv
中文介绍 Code2LoRA 提出用超网络生成 LoRA 适配器,以应对代码语言模型在软件演化中的知识更新问题。无需为每个仓库微调,通过超网络从代码上下文动态生成适配器权重,显著降低仓库级部署成本并提升对新版本的适应性。
👍 45
06/04 08:00
Role-playing language agents (RPLAs) should play characters whose values and behavior evolve as the story progresses, not maintain a fixed persona. Existing benchmarks measure factual recall at a given chapter, not whether responses align with the character's psychological trajectory, especially in
中文介绍 ArcANE 评估角色扮演语言代理(RPLA)能否随剧情发展动态调整角色价值观和行为,而非保持固定人格。现有基准仅测试事实回忆,ArcANE 通过构建故事弧线中的心理轨迹对齐任务,揭示模型在角色演化一致性上的不足。
👍 38
06/03 08:00
Agents are widely deployed as assistants over documents, tools, and code. However, they typically act only on explicit user requests, which surface only the problems the user has noticed, while many other important problems coexist, hidden in plain sight, within the broader user context, with their
中文介绍 TIDE 提出模板引导的迭代式主动发现问题框架,使代理能超越用户显式请求,在文档、工具和代码等上下文中自主发现并报告多个隐藏问题。通过模板化提示和迭代搜索,提升代理的主动性和问题覆盖范围。
👍 37
06/04 08:00
Planning for real-world problems by language models often involves both world and user constraints, which may not be fully specified upfront and are progressively disclosed through interaction. However, existing benchmarks still underexplore adaptive planning under such progressively revealed dual c
中文介绍 AdaPlanBench 评估大语言模型代理在逐步披露的世界约束和用户约束下的自适应规划能力。现有基准缺乏对双重约束渐进揭示场景的测试,该工作设计了包含动态约束变化的规划任务,量化模型在交互中调整计划的效果。
👍 34
06/03 08:00
We introduce VideoKR, the first large-scale training corpus specifically designed to strengthen knowledge- and reasoning-intensive video understanding. It comprises 315K video reasoning examples over 145K newly collected, CC-licensed, expert-domain videos. We develop a human-in-the-loop, skill-orien
中文介绍 VideoKR 构建首个大规模训练语料库,用于强化知识推理型视频理解。包含 31.5 万视频推理样本和 14.5 万专家领域视频,采用人机协作的技能导向标注流程。相比现有基准,在知识密集型视频问答任务上显著提升准确率。
👍 24
06/02 08:00
While household robots are often evaluated based on task completion, everyday domestic environments involve value-conflicting situations in which robots are expected to choose actions that prioritize other values than task success, such as human autonomy, efficiency, or social appropriateness. Yet,
中文介绍 RobotValues 提出评估家庭机器人在人类价值观冲突场景中的行为选择框架。传统指标只关注任务完成,该工作引入价值冲突情境(如自主性、效率、社交恰当性),测试机器人是否优先考虑非任务价值,揭示现有模型的局限。
👍 23
06/04 08:00
Prior work has shown that large language models (LLMs) can translate unseen or low-resource languages by undergoing continued training or even by encoding a grammar book in their context. However, both methods typically overfit specific languages, with limited zero-shot transfer at test time. To tra
中文介绍 针对未见语言翻译任务,提出强化学习(RL)引导模型在上下文中学习新语言翻译。相比继续训练或编码语法书,该方法通过 RL 优化上下文学习策略,实现更好的零样本跨语言迁移,且不依赖特定语言过度拟合。
👍 21
06/04 08:00
Developing unified video generation and editing models capable of interpreting interleaved multimodal inputs is a promising yet challenging frontier field. Existing unified frameworks predominantly rely on massive models (typically 13B parameters or more) and incorporate source video conditions for
中文介绍 LoomVideo 提出统一多模态输入的视频生成与编辑框架,支持交错图像、文本、视频等条件。不同于依赖 13B+ 参数大模型的方法,该工作以更小规模模型实现高质量视频生成与编辑,在多项指标上达到或超越现有方案。
👍 19
06/03 08:00
Experience internalization converts contextual experience from past interactions into reusable parametric capability, offering a promising path toward continual learning in large language models (LLMs). While prior work has predominantly focused on single-iteration transfer, we discover that under m
中文介绍 重新思考持续经验内化:发现多轮迭代比单次迁移更能将交互经验转化为 LLM 的可复用参数化能力。提出多轮内化策略,通过逐步强化关键经验,提升自进化代理在长序列任务中的性能。
👍 19
06/03 08:00
We study the personal camera roll visual question answering setting. In this setting, a conversational AI assistant can access a user's personal camera roll and retrieve relevant photos to answer queries, ranging from simple factual questions (e.g., ``Name of the food I tried yesterday?'') to more o
中文介绍 研究个人相机胶卷的视觉问答设定,对话 AI 可访问用户相册检索相关照片回答简单事实或复杂推理问题。提出端到端检索-推理框架,结合多模态嵌入和场景理解,在真实照片问答上取得高准确率。
👍 15
06/04 08:00
Video generation models have made impressive strides in synthesizing visually compelling content, yet their outputs remain confined to the virtual domain. A natural question follows: how well do these models reflect the physical world when their generated videos leave the screen and enter reality? W
中文介绍 Dream.exe 探究视频生成模型能否「梦想」出可执行的机器人操作。将生成的逼真操作视频转化为真实机器人控制策略,提出闭环仿真-现实迁移方法,验证生成模型在物理世界中的可行性,在多项抓取任务上成功率超 70%。
👍 13
06/04 08:00
Large language model (LLM) agents are increasingly applied to long-horizon tasks such as scientific discovery and machine learning engineering (MLE), where sustained self-evolution becomes a key capability. However, existing MLE agents suffer from inter-branch information isolation, memoryless searc
中文介绍 MLEvolve 提出自进化框架用于自动化机器学习算法发现。现有代理存在分支间信息隔离、搜索无记忆等问题;该框架通过结构化记忆和进化搜索策略,在算法搜索效率和发现质量上显著提升,自动找到优于人工设计的优化器。
👍 11
06/04 08:00
Inference-time skill augmentation provides a lightweight way to improve data-analytic agents by injecting reusable procedural knowledge without updating model parameters. However, discovering effective skills for data analysis remains challenging, as reliable supervision is expensive and success cri
中文介绍 无监督技能发现方法用于增强数据分析代理。无需昂贵监督,通过自生成执行轨迹和成功准则,自动提取可复用的过程性技能。注入这些技能后,代理在复杂数据分析任务上的完成率提升 30% 以上。
👍 10
06/04 08:00
Vision-Language-Action (VLA) models leverage the rich world knowledge of pretrained vision-language models (VLMs) to enable instruction-following robotic manipulation. However, the structural mismatch between VLM semantic spaces and embodied control policies often hinders the learning of precise per
中文介绍 AffordanceVLA 提出视觉-语言-动作模型,通过可操作性感知理解增强动作生成。解决 VLM 语义空间与机器人控制策略的结构错配,显式建模物体可操作区域,在真实机器人指令跟随操作任务中操作成功率提升 15% 以上。
👍 9
06/02 08:00
Selection is a core operation in interactive image editing. To be practical, a user should be able to specify and disambiguate the desired selection region through either text or click-based interactions, and the system should support selecting not only objects but also other criteria, such as mater
中文介绍 MAOAM 提出统一的对象和材质选择方法,用户可通过文本或点击交互指定选择区域,同时支持对象和材质等不同语义标准。基于视觉-语言模型实现灵活解耦选择,在多种编辑任务中取得优于传统交互分割的精度。
👍 8
06/04 08:00
Large language models can reproduce training data, but existing memorization evaluations mostly measure whether models can be forced to do so, rather than whether they do so under ordinary use. We introduce PropMe, a propensity-aware framework for memorization evaluation that contrasts prefix-based
中文介绍 PropMe 引入倾向性感知的记忆化评估框架,区分「能被强迫」泄露与「会主动」泄露训练数据。通过对比前缀强制生成与自发生成,发现模型在普通使用中记忆泄露风险远低于强制评估结果,提供更贴近实际的度量。
👍 7
06/02 08:00
Inference-time scaling has emerged as a critical avenue for enhancing Large Language Models' performance, yet real-world deployment is constrained by strict computational budgets. In this work, we formulate inference budget allocation as a global constrained optimization problem governed by economic
中文介绍 从经济学视角优化 LLM 推理预算分配:形式化为全局约束优化问题,引入「影子价格」概念衡量每单位计算资源的边际收益。实验表明,合理分配推理计算可在固定预算下提升 20% 以上的性能,指导实际部署取舍。
👍 7
06/04 08:00
We propose world-language-action (WLA) models as a new class of embodied foundation models. WLA takes textual instructions, images, and robot states as inputs to jointly predict textual subtasks, subgoal images, and robot actions, conjoining the world modeling interface to learn from extensive egoce
中文介绍 提出世界-语言-动作模型(WLA),统一世界建模、语言推理和动作合成。以文本指令、图像和机器人状态为输入,联合预测子任务描述、子目标图像和机器人动作,在多样化操作任务中实现零样本泛化,成功率远超先前方法。
👍 7
06/04 08:00
Large language models often improve reasoning by generating explicit chain-of-thought (CoT), demonstrating the importance of intermediate computation. However, textual CoT forces this computation through a discrete, serial, and communication-oriented token stream: each reasoning step must be verbali
中文介绍 提出潜在推理归一化流(Latent Reasoning with Normalizing Flows),替代离散文本链式思维。在连续潜在空间中执行隐式推理步骤,避免串行化 token 流约束。在数学推理等任务上以更少计算步骤达到与 CoT 相当甚至更高的准确率。
👍 6
05/28 08:00
Memory-augmented LLM agents tackle complex long-horizon tasks by recursively summarizing interaction trajectories into compact memory. However, existing approaches typically train these memory policies using outcome-based reinforcement learning, failing to localize where intermediate memory quality
中文介绍 元认知记忆策略优化(Meta-Cognitive Memory Policy Optimization)用于长序列 LLM 代理。现有方法用结果奖励训练记忆策略,无法定位中间记忆质量问题;该工作引入过程级奖励,显式优化记忆的准确性和有用性,显著提升长任务成功率。