Daily Papers · 2026-10-13
📄 论文日报 | 来源：Hugging Face Daily Papers
🗓 Episode：2026.10.12 | 扩散模型RL微调高效稳定；多模态模型空间工作区
1. DiffusionART：高效、稳定且可扩展的扩散模型强化学习
- 英文标题：DiffusionART: Efficient, Stable and Scalable Reinforcement of Diffusion Models
- arXiv：2610.12572
💡 一句话解读：扩散模型用强化学习微调时，现有方法要么效率低（反向过程方法），要么不稳定（前向过程方法需调EMA）。本文提出DiffusionART，把每个提示的奖励聚合成一个共享的改进方向，并加入信任域掩码的二次更新，无需EMA调参，训练高效稳定，在图像和视频生成上大幅节省计算资源。
在线强化学习（RL）利用黑盒奖励微调扩散模型，但Flow-GRPO等反向过程方法因信用分配不佳而效率受限，而DiffusionNFT等前向过程方法虽高效，却因需要精心调优的指数移动平均（EMA）离策略训练而不稳定。我们发现NFT风格的目标函数包含一种隐藏的前向-反向一致性力，联合正负训练只能部分抵消，而EMA通过恢复而非离策略 rollout 来稳定训练。我们提出DiffusionART（聚合优先奖励倾斜的扩散强化），它将每个提示的奖励聚合为反向过程状态的一个共享的奖励改进方向，从而在构造上消除一致性力，并添加信任域掩码的第二次更新，实现漂移自适应恢复而无需EMA调优。该方法不需要可微奖励模型，在每提示样本更少时仍有效，并支持成对奖励模型（DiffusionART-Pairwise）。在图像和视频生成中，DiffusionART无需逐任务调参即可高效稳定地训练。特别是在HunyuanVideo-13B上，使用等权重视觉和运动质量奖励时，DiffusionART达到DiffusionNFT和Flow-GRPO-Fast的后期平均奖励水平，GPU小时数分别减少12.02倍和11.72倍，同时平均训练奖励分别高出28.8%和53.0%。
📄 English Abstract
Online reinforcement learning (RL) fine-tunes diffusion models with black-box rewards, but reverse-process methods such as Flow-GRPO are limited in efficiency by poor credit assignment, while forward-process methods such as DiffusionNFT are efficient yet unstable without carefully tuned exponential-moving-average (EMA) off-policy training. We show that NFT-style objectives contain a hidden forward–reverse consistency force that joint positive–negative training cancels only partially, and that EMA stabilizes training through restoration rather than off-policy rollouts. We introduce DiffusionART (Diffusion Reinforcement with Aggregation-First Reward Tilting), which aggregates each prompt's rewards into a shared reward-improving direction for reverse-process states, removing the consistency force by construction, and adds a trust-region-masked second update for drift-adaptive restoration without EMA tuning. It requires no differentiable reward model, remains effective with fewer samples per prompt, and supports pairwise reward models (DiffusionART-Pairwise). Across image and video generation, DiffusionART trains efficiently and stably without per-task tuning. In particular, on HunyuanVideo-13B with equally weighted visual and motion quality rewards, DiffusionART reaches the late-stage mean reward levels of DiffusionNFT and Flow-GRPO-Fast with 12.02 and 11.72 times fewer GPU hours, respectively, while achieving 28.8% and 53.0% higher mean training rewards.
2. S-Space：探索多模态语言模型中的空间工作区
- 英文标题：S-Space: Exploring Spatial Workspace in Multimodal Language Models
- arXiv：2610.13527
💡 一句话解读：多模态大模型能回答空间问题，但内部怎么做到的并不清楚。本研究发现了模型内部一个叫S-Space的“空间工作区”：物体信息在这里被编码成上下、左右、远近的连续坐标，感知把位置写进去，推理能改写它，回答时再读取。作者还用它诊断出模型空间推理的短板——能记住位置，却不擅长换视角做变换。
多模态语言模型（MLLMs）展现出有前景的空间推理能力，但支撑这种能力的内部机制仍鲜为人知。我们发现了S-Space，这是一个线性表示子空间，其中物体词元（object-token）表示沿水平、垂直和距离轴编码连续坐标。尽管我们是通过对物体词元激活求空间答案分数的微分来识别这些轴的，但S-Space是内在的，而非由特定数据或提示诱导产生，并且在非空间问题下仍可读取。我们表明，S-Space充当MLLMs的内部空间工作区：感知将物体位置编码进其中，形成可直接读出以支持准确空间判断的空间表示；内部推理会为视角变化而变换它；言语回答则有选择地访问它——这一点由因果干预所证明，干预会改变空间判断，却在很大程度上保留非空间回答。案例研究进一步表明，S-Space会为未见过的实体和观察者分配位置，并将物理方向与抽象语言含义联系起来，这既说明大规模语言训练的能力，也说明其局限。最后，我们将S-Space用作诊断空间推理失败的工具：方向词读出揭示了基本方位与图像相对参考系之间的系统性耦合；更重要的是，一种感知-计算分解暴露出一个关键瓶颈：MLLMs能够形成有能力的空间表示，却在跨视角变化进行可靠推理所需的空间变换方面有所不足。
📄 English Abstract
Multimodal language models (MLLMs) exhibit promising spatial reasoning capabilities, yet the internal mechanisms supporting them remain poorly understood. We uncover S-Space, a linear representation subspace in which object-token representations encode continuous coordinates along horizontal, vertical, and distance axes. Although we identify these axes by differentiating spatial-answer scores with respect to object-token activations, S-Space is intrinsic rather than induced by specific data or prompts, remaining readable under non-spatial questions. We show that S-Space functions as an internal spatial workspace of MLLMs. Perception encodes object locations into it, forming spatial representations that can be directly read out to support accurate spatial judgments. Internal reasoning transforms it for changing viewpoints. Verbal responses selectively access it, evidenced by causal interventions that alter spatial judgments while largely preserving non-spatial responses. Case studies further show that S-Space assigns locations to unseen entities and the observer, and connects physical directions with abstract linguistic meanings, suggesting both the capabilities and limitations of large-scale language training. Finally, we use S-Space as a diagnostic tool for spatial reasoning failures. Direction-word readouts reveal systematic coupling between cardinal and image-relative reference frames. More importantly, a perception-computation decomposition exposes a key bottleneck: MLLMs can form capable spatial representations, yet fall short in the spatial transformations required to reason reliably across viewpoint changes.
3. Reflex：视觉接地的反应式人形机器人控制
- 英文标题：Reflex: Visually Grounded Reactive Humanoid Control
- arXiv：2610.12772
💡 一句话解读：人形机器人想接住朝自己飞来的箱子，需要全身协调、实时反应，还要靠自身摄像头看懂物体运动。现有方法大多只解决其中一部分。本文提出 Reflex 框架，分三步训练：先学全身接物，再学从延迟不完整的观察中推断箱子运动，最后让视觉系统直接从第一视角图像恢复这种运动信息。在仿真中成功率达85.7%，真机上也成功接住了人扔的箱子。
人类能毫不费力地对周围世界做出反应：他们能娴熟地回击网球、接住掉落的杯子、闪避经过的推车，或在卡车卸货时接住抛来的箱子。这些日常动作需要预判移动目标，并在约一秒内以全身做出反射式反应。这种视觉反应式全身控制对人形机器人而言仍具挑战，因为它们必须（1）协调全身控制，（2）实时响应物体动态，（3）将感知建立在第一视角（egocentric）视觉之上。我们以接住抛来的箱子作为代表性任务，它同时要求上述三种能力；现有系统仅覆盖其中一部分，依赖末端执行器、动作捕捉或带标记的物体。我们提出 Reflex，一个三阶段框架，逐阶段学习这些能力，并通过物体动态的紧凑表示将它们连接起来。首先，基于状态的策略通过强化学习学会全身接物。其次，该策略学会从延迟且不完整的观测中推断移动箱子的动态。第三，第一视角 RGB-D 编码器学会直接从视觉观测中恢复这种动态表示，同时保持全身控制器固定。这种分解为每种能力提供了直接的训练信号，并使视觉学习更具可扩展性。在仿真中，Reflex 仅凭第一视角 RGB-D 达到 85.7% 的接物成功率，而给定特权箱子观测时为 87.9%。时序视觉历史至关重要：去除它会使居中投掷和侧向偏移投掷的成功率分别下降 35.6 和 39.8 个百分点。在真实的宇树 G1 机器人上，Reflex 接住人类抛出的箱子，在 2.5 米和 5 米距离处成功率分别为 65% 和 40%，达到人类接物者在相同距离下成功率的 76% 和 59%。
📄 English Abstract
Humans effortlessly react to the world; they adeptly return a tennis shot, catch a falling cup, step aside from a passing cart, or pass a box tossed during truck unloading. These everyday actions require anticipating a moving target and reflexively responding with the whole body within about a second. Such visual reactive whole-body control remains challenging for humanoids, which must (1) coordinate whole-body control, (2) react to object dynamics in real time, and (3) ground perception in egocentric vision. We study catching thrown boxes as a representative task that demands all three capabilities; existing systems cover only part of it, relying on end-effectors, motion capture, or marked objects. We present Reflex, a three-stage framework that learns these capabilities stage by stage while connecting them through a compact representation of object dynamics. First, a state-based policy learns whole-body catching with reinforcement learning. Second, the policy learns to infer the dynamics of a moving box from delayed and incomplete observations. Third, an egocentric RGB-D encoder learns to recover this dynamics representation directly from visual observations while keeping the whole-body controller fixed. This decomposition provides each capability with a direct training signal and makes visual learning substantially more scalable. In simulation, Reflex achieves 85.7% catch success from egocentric RGB-D, compared with 87.9% when given privileged box observations. Temporal visual history is critical: removing it reduces success by 35.6 and 39.8 points on centered and laterally offset throws. On a real Unitree G1, Reflex catches human-thrown boxes with 65% and 40% success at 2.5 and 5 m, reaching 76% and 59% of human catchers' success at the same distances.
4. SoftwareGym：利用多样化合成软件环境扩展计算机使用智能体训练
- 英文标题：SoftwareGym: Scaling Computer-Use Agent Training with Diverse Synthesized Software Environments
- arXiv：2610.12611
💡 一句话解读：训练能操作电脑的AI助手时，缺少大量真实又多样的软件环境，导致训练效果差、容易过拟合。本文提出SoftwareGym，自动生成各种可运行的软件界面，并反复检查修复，用这些合成环境训练AI，使其在真实电脑操作测试中表现明显提升。
计算机使用智能体（CUAs）正在快速进步，但大规模合成多样且可靠的软件环境仍具挑战。原生应用环境需要大量构建与集成工作，限制了应用覆盖范围并增加了针对特定应用过拟合的风险；而浏览器托管环境则交互深度和工作流复杂度有限。这些局限促成了SoftwareGym——一个自动化的流水线，用于合成具有多样界面、可执行功能和运行时验证行为的图形用户界面（GUI）软件环境。它生成结构多样的设计，将其实现为可运行应用，并通过GUI评估与迭代修复来确保功能可靠性。利用SoftwareGym，我们合成了涵盖12个类别的182个应用，平均GUI任务通过率达到91.5%。我们构建了SoftwareGym-Bench，一个包含120个任务、覆盖相同12个类别的评估套件，并利用从这些环境中收集的轨迹进行CUA训练。所得模型在严格域外的OSWorld基准上，将Pass@3从41.39%提升至56.97%，并在SoftwareGym-Bench上达到40.0%。这些结果展示了合成软件环境作为可扩展且可迁移的监督信号用于CUA训练的潜力。
📄 English Abstract
Computer-use agents (CUAs) are rapidly improving, yet synthesizing diverse and reliable software environments at scale remains challenging. Native application environments require substantial construction and integration effort, which limits application coverage and increases the risk of application-specific overfitting. Browser-hosted environments provide limited interaction depth and workflow complexity. These limitations motivate SoftwareGym, an automated pipeline for synthesizing GUI software environments with diverse interfaces, executable functionality, and runtime-verified behavior. It generates structurally diverse designs, realizes them as runnable applications, and uses GUI evaluation with iterative repair to ensure functional reliability. Using SoftwareGym, we synthesize 182 applications across twelve categories and achieve a mean GUI task pass rate of 91.5%. We construct SoftwareGym-Bench, a 120-task evaluation suite spanning the same twelve categories, and use trajectories collected from these environments for CUA training. The resulting model improves Pass@3 from 41.39% to 56.97% on the strictly out-of-domain OSWorld benchmark and achieves 40.0% on SoftwareGym-Bench. These results demonstrate the potential of synthesized software environments as scalable and transferable supervision for CUA training.
5. EmbodiedCLBench：评估自进化具身智能体的持续学习
- 英文标题：EmbodiedCLBench: Evaluating Continual Learning for Self-Evolving Embodied Agents
- arXiv：2610.13021
💡 一句话解读：现有持续学习研究大多停留在非具身环境，无法评估智能体在真实模拟场景中积累经验、解决新任务的能力。为此，作者提出首个具身持续学习基准EmbodiedCLBench，让智能体先学基础任务，再重组技能解决未见复杂任务，并测试多种方法，发现经验能稳定提升表现，但样本增多效果不增。
持续学习要求模型积累过往经验并将其迁移到新任务。尽管持续学习已有大量研究，但现有工作大多局限于非具身（non-embodied）环境，尚未扩展到具身环境。为填补这一空白，我们提出了EmbodiedCLBench，这是首个用于评估自进化智能体持续学习能力的具身模拟基准。该基准以组合泛化（compositional generalization）为核心评估原则，检验智能体能否先从基础任务中学习，再重组已获得的技能以解决复杂、未见的高级任务。基于EmbodiedCLBench，我们在多种智能体框架和学习方法上开展了大量实验，揭示了持续存在的局限性和独特行为。结果表明，虽然高级任务在零样本（zero-shot）设置下几乎无法解决，但基础任务的经验能带来持续改进；即使任务复杂度增加，这种经验带来的收益也保持稳定。然而，这种改进并不随学习样本数量的增加而相应扩展，当前智能体无法从额外经验中提取有意义的信息。总体而言，我们的发现凸显了持续学习是一项基础但尚未被充分探索的能力，而我们的基准为在具身环境中设计有效的自进化智能体提供了宝贵资源。代码和数据集见https://pku-value-lab.github.io/EmbodiedCLBench-homepage。
📄 English Abstract
Continual learning requires models to accumulate past experience and transfer it to new tasks. Despite substantial research on continual learning, existing work remains largely confined to non-embodied settings, without extending to embodied environments. To address this gap, we introduce EmbodiedCLBench, the first benchmark in embodied simulation for evaluating the continual learning capability of self-evolving agents. With compositional generalization as the core evaluation principle, our benchmark evaluates whether agents can first learn from basic tasks and then recombine the acquired skills to solve complex, unseen advanced tasks. Based on EmbodiedCLBench, we conduct extensive experiments across multiple agent harnesses and learning methods, revealing consistent limitations and distinctive behaviors. Our results show that while advanced tasks can be hardly solved under zero-shot setting, the experience from basic tasks enables consistent improvement. Such benefit from the experience remains stable even if the task complexity increases. However, this improvement does not scale accordingly with the number of learning samples, and current agents cannot extract meaningful information from additional experience. Collectively, our findings highlight continual learning as a fundamental yet underexplored capability, and our benchmark offers valuable resources for designing effective self-evolving agents in embodied environments. Our code and dataset are available at https://pku-value-lab.github.io/EmbodiedCLBench-homepage.
6. 用立体声聆听世界：学习动态空间对应以实现沉浸式音视频联合生成
- 英文标题：Hear the World in Stereo: Learning Dynamic Spatial Correspondence for Immersive Joint Video-Audio Generation
- arXiv：2609.38748
💡 一句话解读：现有音视频生成模型能同步画面和声音，但无法生成随画面移动而改变方位的立体声。本文提出 StereoBind，用运动轨迹把画面中物体的移动和立体声方位绑定，并构建了带轨迹的数据集和评测基准，让生成的声音方位与画面运动一致。
近期的联合视频-音频生成模型已实现较强的语义对应和时间同步。然而，AR/VR 和交互式游戏等应用进一步要求立体声以提供沉浸感，这一点在很大程度上被忽视。有效的立体声要求感知到的声音位置与其对应视觉源的运动持续一致。我们将这一性质称为动态空间对应（Dynamic Spatial Correspondence），并提出 StereoBind，一个将视觉源运动与立体声生成绑定的框架。StereoBind 通过三种互补机制利用运动轨迹来协调视觉运动和立体声：视觉运动绑定（Visual Motion Binding）建立源感知的音视频对应，空间轨迹编码器（Spatial Track Encoder）捕捉绝对源位置，残差轨迹 RoPE（Residual Track RoPE）建模相对运动。为了监督和评估，我们构建了 StereoWorld-29K，一个带有配对运动轨迹的大规模立体声音视频数据集，以及 StereoWorldBench，用于衡量音视频空间一致性。实验表明，StereoBind 在保持整体音视频质量的同时，显著提升了立体声生成中的空间对齐效果，优于现有模型。
📄 English Abstract
Recent joint video-audio generation models have achieved strong semantic correspondence and temporal synchronization. However, applications such as AR/VR and interactive gaming further require stereo audio to provide an immersive sense, which remains largely overlooked. Effective stereo audio requires the perceived sound location to evolve consistently with the motion of its corresponding visual source. We refer to this property as Dynamic Spatial Correspondence and propose StereoBind, a framework that binds visual source motion to stereo sound generation. StereoBind uses motion tracks to coordinate visual motion and stereo audio through three complementary mechanisms. Visual Motion Binding establishes source-aware audiovisual correspondence, the Spatial Track Encoder captures absolute source positions, and Residual Track RoPE models relative motion. For supervision and evaluation, we construct StereoWorld-29K, a large-scale stereo audio-video dataset with paired motion tracks, and StereoWorldBench for measuring audiovisual spatial consistency. Experiments show that StereoBind substantially improves spatial alignment in stereo audio generation over existing models while preserving overall audiovisual quality.
7. VisionWeave：将弹性视觉表征编织为多模态大语言模型的原生能力
- 英文标题：VisionWeave: Weaving Elastic Visual Representations as a Native Capability of MLLMs
- arXiv：2610.07987
💡 一句话解读：多模态大模型处理图片时，不管内容重要与否都切成同样多的碎片，既浪费算力又丢失细节。VisionWeave 让模型自己学会哪里该细看、哪里可粗略表示，从而在几乎不掉性能的前提下省下大量计算。
多模态大语言模型（MLLM）已成为视觉理解的主流范式，但其将输入编码为稠密、固定尺寸的图像块（patch）token 的方式带来了巨大开销。然而视觉信息分布并不均匀：部分区域需要细粒度细节，另一些区域则可采用紧凑表示。下采样会牺牲细节，而现有的 token 剪枝与自适应方法在内容自适应粒度、任务泛化性以及与当代 MLLM 和服务基础设施的集成方面仍存在局限。为克服这些限制，需要基础模型能够端到端地学习在何处、以何种粒度分配视觉表示，我们将这一原生能力称为弹性视觉表征编织（elastic visual representation weaving）。我们提出 VisionWeave，通过大规模训练在前沿级 MLLM 中建立该能力。它包含两个组件：门控空间池化器（gated spatial pooler）在共享的 MRoPE 坐标下，与原生细粒度表示一同构建粗粒度表示；粒度路由器（granularity router）则学习二者的内容自适应分配。仅通过自蒸馏，我们在 Qwen3.5-4B 上验证了该能力，并扩展至 Qwen3.8-27B，消耗超过 3 万 A100 GPU 小时。基于 Qwen3.8-27B，VisionWeave 能根据视觉内容自适应调整 token 节省量，在八个基准上平均节省 43.0% 的 token，同时保留 98.9% 的原生性能；相比之下，固定 50% 节省目标的 token 剪枝基线仅保留 88% 的性能。大量评估证实，该方法在不同任务、分辨率和视频帧数下均能实现稳健的效率-质量权衡。在 SGLang 服务引擎上部署时，我们的方法实现了 2.3 倍的吞吐量提升，同时将平均首 token 时延（TTFT）降低 54.4%，平均每 token 输出时延（TPOT）降低 60.6%。我们相信，这些结果使弹性视觉编织成为下一代多模态模型的一项有前景的能力。
📄 English Abstract
Multimodal large language models have become the dominant paradigm for visual understanding, but incur substantial costs by encoding inputs into dense, fixed-size patch tokens. However, visual information is unevenly distributed: some regions require fine-grained detail, while others admit compact representations. Downsampling sacrifices this detail, while existing token pruning and adaptive approaches remain limited in content-adaptive granularity, task generalization, and integration with modern MLLMs and serving infrastructure. Overcoming these limitations calls for foundation models that learn, end to end, where-and at what granularity-to allocate visual representations, a native capability we term elastic visual representation weaving. We introduce VisionWeave, establishing this capability in frontier-level MLLMs through large-scale training. It combines two components: a gated spatial pooler constructs coarse-grained representations alongside native fine-grained representations within a shared MRoPE coordinate, while a granularity router learns their content-adaptive allocation. Through self-distillation alone, we validate this capability on Qwen3.5-4B and scale to Qwen3.8-27B with over 30K A100 GPU-hours. Based on Qwen3.8-27B, VisionWeave adaptively adjusts token savings to visual content, saving 43.0% tokens on average while retaining 98.9% native performance across eight benchmarks, versus only 88% performance preserved for token pruning baselines with a fixed 50% savings target. Extensive evaluations confirm robust efficiency-quality trade-offs across diverse tasks, resolutions and video frames. When deployed on SGLang serving engine, our method achieves a 2.3x throughput gain while reducing mean TTFT by 54.4% and mean TPOT by 60.6%. Together, we believe these results position elastic visual weaving as a promising capability for next-generation multimodal models.
8. ProgramGym：扩展长时程程序复现智能体的训练环境
- 英文标题：ProgramGym: Scaling Training Environments for Long-Horizon Program Reproduction Agents
- arXiv：2610.13395
💡 一句话解读：现有编程智能体多用于修 bug 或按需求文档生成代码，缺少能让它像工程师一样从零写程序、自己摸索需求并做设计的训练环境。作者提出 ProgramGym，用自动修复的沙箱和黑盒测试批量造出 1648 个可验证训练环境，并设计分析、实现、审查分工的智能体流程，显著提升了程序复现能力。
编程智能体正越来越多地被用于像软件工程师那样从零构建程序，而非解决仓库内的问题或根据给定需求文档生成仓库。然而，能够迫使智能体自行发现需求并决定系统应如何设计的、可扩展且可验证的训练环境仍然稀缺。此外，现有智能体工作流也无法支撑长时程开发——即智能体在实现之前先探索参考程序。为应对这些挑战，我们提出 ProgramGym，一个由 Program-Env 和 Program-Agent 组成的框架，用于构建长时程程序复现智能体的可扩展训练环境。具体而言，Program-Env 通过一个反馈驱动的循环，修复运行时沙箱构建和黑盒测试的质量，从而从代码仓库构建可验证的训练环境。借助这一新流程，我们获得了覆盖 C、C++、Go 和 Rust 的 1,648 个训练环境。Program-Agent 将分析、实现和审查的角色隔离工作流与用于编译检查和失败恢复的修订管理相结合。在 ProgramBench 上的实验表明，Program-Env 和 Program-Agent 均提升了程序复现的性能。在这些构建环境生成的轨迹上进行微调，将 Qwen3.8-27B 的性能从 34.87% 提升至 42.84%。Program-Agent 进一步将基础模型和 SFT 模型分别提升至 55.71% 和 56.25%，均超过部分旗舰模型。我们将发布 ProgramGym，以支持关于从零构建程序的编程智能体的进一步研究。
📄 English Abstract
Coding agents are increasingly used to build programs from scratch as software engineers do rather than to resolve issues within a repository or to generate a repository from a given requirements document. However, scalable verifiable training environments that force agents to discover what is required and to decide how the system should be designed remain scarce. Moreover, existing agent workflows also cannot sustain long-horizon development in which the agent explores the reference program before implementation. To address these challenges, we introduce ProgramGym, a framework consisting of Program-Env and Program-Agent for scalable training environments of long-horizon program reproduction agents. Specifically, Program-Env constructs verifiable training environments from repositories through a feedback-driven loop that repairs runtime sandbox builds and the quality of black-box tests. Using this novel pipeline, we obtain 1,648 training environments spanning C, C++, Go, and Rust. Program-Agent combines a role-isolated workflow of analysis, implementation, and review with revision management for compilation checks and failure recovery. Experiments on ProgramBench show that both Program-Env and Program-Agent improve performance on program reproduction. Fine-tuning on trajectories with these constructed environments raises the performance of Qwen3.8-27B from 34.87% to 42.84%. Program-Agent further raises the base and SFT models to 55.71% and 56.25%, both exceeding some flagship models. We will release the ProgramGym to support further research on coding agents that build programs from scratch.
9. Work2Skill：捕获人类 GUI 经验以实现可复用智能体技能归纳
- 英文标题：Work2Skill: Capturing Human GUI Experience for Reusable Agent Skill Induction
- arXiv：2610.12984
💡 一句话解读：电脑操作智能体遇到不熟悉的软件时常常不会用，因为缺少人类经验。本文提出 Work2Skill 框架，先记录人类操作屏幕、鼠标键盘等数据，再从中提炼出可复用的操作规则和步骤，让智能体学会后能更好地完成任务。
计算机使用智能体（computer-use agents）可以执行日益复杂的图形用户界面（GUI）任务，但面对不熟悉的软件和需要人类领域知识的工作流程时仍具挑战性。人类演示通过相同的 GUI 交互空间传递过程性知识，但复用这些知识需要在上下文中解释动作，并将工作流规则与任务特定细节分离。为此，我们提出了 Work2Skill 框架，用于捕获人类 GUI 经验以归纳可复用的智能体技能。WorkRecorder 为该框架提供了收集此类经验的基础设施，能够对齐屏幕录制、鼠标和键盘事件以及可用的无障碍树（accessibility-tree）信息。利用 WorkRecorder，我们构建了 WorkDemo-350，这是首个用于可复用智能体技能归纳及下游评估的同步多模态人类桌面 GUI 演示数据集，其 350 个演示均经过完整性和任务成功性筛选。随后，我们提出了一种智能体式技能归纳方法，该方法按需检查证据，以生成捕获工作流规则、步骤依赖关系和完成检查的可复用技能。在相同的执行设置下，当四个相同的执行模型分别配备由各方法归纳出的技能时，Work2Skill 取得了最高分数和成功率。相较于对应的无技能基线，在 MyPCBench Bounded Action 上平均得分提升 4.67–10.54 分，在 OSWorld-Verified 的 286 个任务子集上提升 3.11–7.50 分。我们的代码和数据可在 https://github.com/china-qijizhifeng/Work2Skill 获取。
📄 English Abstract
Computer-use agents can perform increasingly complex GUI tasks, yet unfamiliar software and workflows requiring human domain knowledge remain challenging. Human demonstrations convey procedural knowledge through the same GUI interaction space, but reuse requires interpreting actions in context and separating workflow rules from task-specific details. We thus introduce Work2Skill, a framework that captures human GUI experience for reusable agent skill induction. WorkRecorder provides the framework's infrastructure for collecting this experience, aligning screen recordings, mouse and keyboard events, and available accessibility-tree information. Using WorkRecorder, we construct WorkDemo-350, the first synchronized multimodal dataset of human desktop GUI demonstrations for reusable agent skill induction and downstream evaluation. Its 350 demonstrations are screened for completeness and task success. We then propose an agentic skill-induction method that inspects evidence on demand to produce reusable skills capturing workflow rules, step dependencies, and completion checks. When the same four execution models are equipped with skills induced by each method under shared execution settings, Work2Skill achieves the highest scores and success rates. Mean score improvements over the corresponding No Skill baseline are 4.67–10.54 points on MyPCBench Bounded Action and 3.11–7.50 points on a 286-task subset of OSWorld-Verified. Our code and data are available at https://github.com/china-qijizhifeng/Work2Skill.
10. SepGen：在单一模型中实现多音轨音视频分离与生成
- 英文标题：SepGen: Multi-Stem Audio-Video Separation and Generation in a Single Model
- arXiv：2610.11361
💡 一句话解读：现有音视频生成模型只能输出混合在一起的音轨，无法单独拿到每个声源的声音，导致无法在新视角下重新定位和渲染。SepGen 在预训练音视频生成模型基础上，一次采样同时输出视频、混合音轨以及每个带文字描述声源的独立波形，并支持生成与分离两种模式。
一个四维（4D）音视频场景包含一段视频、视频所描绘的动态几何结构，以及其中分布的各声音源，每个声源都有自己的位置和运动轨迹。要从新视角渲染这样的场景，就要求每个声源都以独立波形的形式存在，这样才能在信号混合之前对其在场景中定位并传播到观察者处。联合音视频生成器可以合成视频及其配乐，但配乐是以单一音频混合的形式输出的，其中各声源无法被单独获取。我们提出 SepGen，它扩展了一个预训练的音视频生成器，使其在一次联合采样过程中同时输出视频、混合音轨，以及每个带描述声源对应的一个波形。SepGen 支持两种互补模式：生成与分离。在生成模式下，每个声源描述指定其对应音轨（stem）的内容，例如一段双人对话会按原始轮流顺序输出每位说话人各自的音轨。在分离模式下，输入音频混合保持干净，而描述指定要提取的内容，因此模型可以根据自由文本描述来分解一段录音。我们在由文本合成的场景上评估生成能力，并在其他生成器渲染的场景以及语音、音乐和音效的真实录音上评估分离能力。给定一段音频混合以及包含台词内容的描述，SepGen 的表现优于语言条件分离器，在语音上尤为明显，并且当描述中移除台词后仍保持领先。代码、检查点和数据集见 https://sepgen.github.io/
📄 English Abstract
A 4D audio-visual scene comprises a video, the dynamic geometry it depicts, and the sound sources that populate it, each with its own position and trajectory. Rendering such a scene from a novel viewpoint requires that every source be available as an individual waveform, so that it can be localized in the scene and propagated to the observer before the signals are mixed. Joint audio-video generators can synthesize the video and its soundtrack, but the soundtrack is emitted as a single audio-mix in which the sources are not individually accessible. We present SepGen, which extends a pretrained audio-video generator to emit the video, the mixed soundtrack, and one waveform per captioned source in one joint sampling run. SepGen supports two complementary modes: generation and separation. In generation mode, each source caption specifies what its stem contains. A two-speaker dialogue, for example, comes out as one stem per speaker in the original turn order. In separation mode, the input audio-mix remains clean while the captions specify what to extract, so the model can decompose a recording from a free-text description. We evaluate generation on scenes synthesized from text, and separation on scenes rendered by other generators and on real recordings of speech, music, and sound effects. Given an audio-mix and captions that carry the spoken lines, SepGen outperforms language-conditioned separators, most clearly on speech, and it keeps the lead when the lines are removed from the captions. Code, checkpoints, and datasets are available at https://sepgen.github.io/
11. 通过合成自蒸馏持续后训练小型模型
- 英文标题：Continually Post-training Small Models via Synthetic Self-distillation
- arXiv：2610.13113
💡 一句话解读：小模型在持续学习新技能时容易忘掉旧知识，而现有的自蒸馏方法依赖大模型才有的强能力，在小模型上失效。作者发现失败根源是训练数据分布偏移，于是提出合成自蒸馏：让一个稍大的同族模型帮忙改写训练数据，再配合筛选，让小模型在不偏离原有分布的情况下学到新东西。
持续后训练面临“获取”与“保持”的双重挑战：模型既要学习新能力，又要保留已有知识。自蒸馏微调（SDFT）通过带上下文示例的自监督缓解了这一问题，但它依赖强大的上下文学习和指令遵循能力，因此在低延迟、端侧使用至关重要的小模型上会失效。为在小规模下实现持续学习，我们首先找出失败的根源，证明获取能力差与遗忘程度同后训练模型的分布偏移呈线性相关，从而揭示分布内学习的必要性。为此我们提出合成自蒸馏（SynthSelf），将 SDFT 的分布内改写任务外包给一个略大的同族对齐模型，并结合离策略监督与拒绝采样，在不破坏小模型原有分布的前提下提供其所需的高质量监督。在 4 个获取数据集（含一个混合数据集）和 14 个保持基准上的评估表明，SynthSelf 在 3 种小模型规模上持续优于 6 种方法（包括离策略与在策略蒸馏）。进一步的大量分析证实，SynthSelf 为有效的小模型持续学习开辟了道路。
📄 English Abstract
Continual post-training faces a dual challenge of acquisition and retention: models must learn new capabilities while preserving prior knowledge. While Self-Distillation Fine-Tuning (SDFT) mitigates this using self-supervision with in-context demonstrations, it relies on strong in-context learning and instruction-following abilities. Consequently, SDFT fails on smaller models critical for low-latency, on-device use. To enable continual learning at smaller scales, we first identify the root cause of failure. We show that poor acquisition and forgetting correlate linearly with the post-trained model's distribution shift, revealing the necessity of in-distribution learning. We thus introduce Synthetic Self-Distillation (SynthSelf), imitating SDFT's in-distribution rewrite by outsourcing the task to a minimally larger, same-family aligner. Coupled with off-policy supervision and rejection sampling, this provides the high-quality supervision small models require without breaking their original distribution. Evaluations spanning 4 acquisition datasets (including a mixture) and 14 retention benchmarks show SynthSelf consistently outperforms 6 methods (including off- and on-policy distillation) across 3 small model scales. Extensive further analyses confirm SynthSelf paves the way toward effective small-model continual learning.
12. AI 智能体能否通过学习登顶？评估长期运行的游戏智能体竞赛中的启发式学习
- 英文标题：Can AI Agents Learn Their Way to the Top? Evaluating Heuristic Learning in a Long-Running Game Agent Competition
- arXiv：2610.12341
💡 一句话解读：这篇研究想弄清：AI 智能体能不能像人类选手一样，通过不断打比赛、看回放来改进自己的游戏策略，最终在长期竞赛中登顶？作者把这种「让 AI 当学习引擎、反复修改可执行策略而模型本身不变」的做法形式化为对抗式启发式学习（AHL），并搭建了 AAArena 基准——包含 12 款真实对抗游戏和 1920 份人类程序，按真实比赛规则评测。结果显示，最强配置拿下 6 枚金牌，但另外 6 款游戏仍无人超越人类，规则越复杂表现越差。
对抗性游戏推动了从启发式搜索到强化学习的进步，但从有限样本中学习和调整策略仍然具有挑战性。AI 智能体提供了另一种思路：把游戏经验转化为对可执行策略的修订。在启发式学习（HL）的基础上，我们形式化提出了对抗式启发式学习（Adversarial Heuristic Learning, AHL）这一范式，它把 AI 智能体当作学习引擎来改进游戏策略及配套软件，同时保持模型权重不变。我们提出了 AAArena 基准，包含 12 款真实对抗性游戏和 1920 份存档的人类程序，并采用模拟真实游戏竞赛的评测协议。智能体需要解读规则、选择对手、分析回放并修改游戏智能体，以在固定的对局与评测预算内取得最高排名。我们评测了多种模型与工具链配置：Opus5.5 搭配 Claude Code 赢得 6 枚金牌，而没有任何受评配置能在其余 6 个人类排行榜上登顶。在规则说明更复杂的游戏中，表现普遍更弱。进一步实验表明，对手选择和密集反馈有助于策略改进，且智能体既能从自身对局的同策略（on-policy）回放中学习，也能从其他玩家对局的异策略（off-policy）回放中学习。这些结果凸显了 HL 在对抗性游戏中的潜力，也指出了在游戏理解、策略实现和长时程策略开发方面持续存在的挑战。
📄 English Abstract
Adversarial games have driven advances from heuristic search to reinforcement learning, yet learning and adapting strategies from limited samples remain challenging. AI agents offer an alternative by turning game experience into revisions of executable policies. Building on heuristic learning (HL), we formalize Adversarial Heuristic Learning (AHL), a paradigm that uses AI agents as learning engines to refine game policies and supporting software while keeping model weights fixed. We introduce AAArena, a benchmark comprising 12 authentic adversarial games and 1,920 archived human programs, with an evaluation protocol modeled on real-world game competitions. Agents interpret rules, choose opponents, analyze replays, and revise game agents to achieve their highest ranking within fixed match and evaluation budgets. We evaluate \val{completedmodels} model and harness configurations: Opus5.5 with Claude Code earns 6 gold medals, while no evaluated configuration tops the remaining 6 human ladders. Performance is generally weaker in games with more complex rule specifications. Further experiments show that opponent selection and dense feedback support policy improvement, and that agents learn from both on-policy replays of their own matches and off-policy replays of other players' matches. These results highlight HL's potential in adversarial games and identify persistent challenges in game understanding, strategy implementation, and long-horizon policy development.
13. SoloQ：面向扩散语言模型的无校准量化
- 英文标题：SoloQ: Calibration-Free Quantization for Diffusion Language Models
- arXiv：2610.07121
💡 一句话解读：扩散语言模型推理成本高，现有量化方法依赖校准数据，但激活分布会随去噪过程变化，导致不准。SoloQ 提出免校准量化：把权重和激活旋转到分布可预测的坐标系，从而无需数据即可量化，还针对块扩散模型压缩缓存，实现省显存、提速。
扩散大语言模型（dLLM）通过基于双向扩散的 token 生成，成为自回归语言模型之外一种有前景的替代方案。然而，其不断增长的模型规模和高推理成本使高效部署面临挑战：全序列去噪会反复调用计算密集的前向传播，而块扩散模型还额外引入内存密集的 KV 缓存。因此，低位权重-激活量化颇具吸引力，但现有 dLLM 训练后量化方法依赖校准数据，尽管激活分布会随掩码状态和去噪步数发生偏移。我们提出 SoloQ，一种免校准量化框架，它将权重和激活映射到具有可预测边缘分布的归一化旋转基中，从而实现与数据无关的量化。SoloQ 将结构化 K-RPBH 旋转与轻量级重缩放校正相结合，以实现免校准量化。其可预测的旋转后分布同时支持分布匹配码本和硬件原生的 NVFP4。对于块扩散模型，SoloQ 进一步采用提交时 KV 缓存量化，在不扰动正在去噪的块的情况下压缩持久状态。在全序列 dLLM（LLaDA 和 Dream）以及块扩散 dLLM（Fast-dLLM v2 和 Nemotron-Labs-Diffusion）上，SoloQ 在 4 比特量化下保持精度，并在知识与推理密集型基准上优于基于校准的基线。使用 NVFP4 时，SoloQ 将峰值内存最多降低 2.61 倍，并将端到端推理最多加速 2.24 倍。
📄 English Abstract
Diffusion large language models dLLMs) have emerged as a promising alternative to autoregressive language models through bidirectional diffusion-based token generation. However, their growing model sizes and high inference costs make efficient deployment challenging: full-sequence denoising repeatedly invokes compute-intensive forward passes, while block-diffusion models additionally introduce a memory-intensive KV-cache. Low-bit weight-activation quantization is therefore attractive, yet existing dLLM post-training quantization methods rely on calibration data despite activation distributions shifting across masking states and denoising steps. We present SoloQ, a calibration-free quantization framework that maps weights and activations into a normalized rotated basis with a predictable marginal distribution, enabling data-independent quantization. SoloQ combines a structured K-RPBH rotation with a lightweight rescaling correction for calibration-free quantization. Its predictable post-rotation distribution supports both distribution-matched codebooks and hardware-native NVFP4. For block-diffusion models, SoloQ further applies commit-time KV-cache quantization to compress persistent states without perturbing the actively denoised block. Across full-sequence dLLMs (LLaDA and Dream) and block-diffusion dLLMs(Fast-dLLM v2 and Nemotron-Labs-Diffusion), SoloQ retains accuracy under 4-bit quantization and outperforms calibration-based baselines on knowledge- and reasoning-intensive benchmarks. With NVFP4, SoloQ reduces peak memory by up to 2.61X and accelerates end-to-end inference by up to 2.24X.
14. PolyOCR-Venus：面向以文本为中心的视觉智能的统一 OCR 基础模型
- 英文标题：PolyOCR-Venus: Unified OCR Foundation Models for Text-Centric Visual Intelligence
- arXiv：2609.37712
💡 一句话解读：现有 OCR 系统通常只擅长单一任务，难以在复杂场景中同时兼顾识别、解析和推理。本文提出 PolyOCR 系列统一模型，用共享的指令框架、大规模数据引擎和一种结合验证与蒸馏的训练方法，让一个模型能应对多种以文本为中心的视觉任务。
光学字符识别（OCR）正从纯文本转录走向通用视觉智能，要求模型在复杂视觉环境中识别、定位并推理文本信息。然而，现有 OCR 系统往往只擅长部分任务，难以在多种场景中平衡识别、解析与推理。本报告提出 PolyOCR，一个不同规模尺度的统一 OCR 基础模型系列。PolyOCR 将共享的指令跟随框架与大规模数据引擎相结合，把异构视觉资源转化为经过质量验证的 OCR 监督数据。我们提出能力引导策略优化（Competence-Guided Policy Optimization），它将基于验证器的组相对策略优化（Group Relative Policy Optimization）与基于教师可靠性和师生能力差距的样本级路由的在线策略蒸馏相结合。我们还提出 OCRBench v2.1，这是我们对 OCRBench v2 的修订版，包含人工核验的标注修正和与任务对齐的评分指标。在 OCRBench v2.1、CC-OCR、内部 KIE Benchmark、OmniDocBench v1.6 和 MDPBench 上的大量实验表明，PolyOCR 达到了最先进或极具竞争力的性能。
📄 English Abstract
Optical Character Recognition (OCR) is evolving from plain-text transcription toward general visual intelligence, requiring models to recognize, localize, and reason over textual information in complex visual environments. However, existing OCR systems often excel at only some tasks and struggle to balance recognition, parsing, and reasoning across scenarios. In this report, we present PolyOCR, a family of unified OCR foundation models of varying scales. PolyOCR combines a shared instruction-following framework with a large-scale data engine that converts heterogeneous visual resources into quality-verified OCR supervision. We introduce Competence-Guided Policy Optimization, which combines verifier-based Group Relative Policy Optimization with on-policy distillation through sample-wise routing based on teacher reliability and the teacher–student competence gap. We also introduce OCRBench v2.1, our revision of OCRBench v2 with manually verified annotation corrections and task-aligned scoring metrics. Extensive experiments across OCRBench v2.1, CC-OCR, in-house KIE Benchmark, OmniDocBench v1.6 and MDPBench demonstrate that PolyOCR achieves state-of-the-art or highly competitive performance.
15. RSI-Anything：用世界模型解锁递归自我改进
- 英文标题：RSI-Anything: Unlocking Recursive Self-Improvement with World Models
- arXiv：2610.13912
💡 一句话解读：这篇研究想解决AI系统如何从经验中持续自我提升，而不用依赖昂贵的外部反馈。它提出一个通用框架，用能预测未来的“世界模型”来指导行动和挑选学习数据，并保留有效的更新，从而在多个领域实现递归式进步。
递归自我改进（RSI）旨在将交互经验转化为持久的能力提升，但外部反馈代价高昂，且随着策略演化，世界模型的预测可能变得不可靠。我们提出RSI-Anything，一个基于JEPA-Anything的领域通用框架，利用预测性世界模型来指导行动和学习数据的选择。观测到的反馈驱动策略或工作流改进以及条件式世界模型自适应，同时组件与兼容性检查控制更新后的策略-模型对的保留。在多种计算与科学领域中，RSI-Anything展现出显著收益：在代码修复中，固定编码器时，成功率从静态验证器的26.3%提升至预测选择的29.2%，再经世界模型自适应后达到31.2%，测试用量相当；图形用户界面（GUI）评估显示任务完成率更高，同时平均操作数减少39.4%；在连续五个策略版本中，实验室规划得分从21.80%翻倍以上至46.01%；在反应优化中，相同40次实验预算下，RSI-Anything达到96.15%的平均最佳观测产率，而贝叶斯优化为82.16%。一个端到端的免疫科学应用进一步将工作流演化与预测建模同治疗假设生成联系起来。这些结果展示了预测性世界模型和保留系统更新如何在不同环境中实现递归改进。
📄 English Abstract
Recursive self-improvement (RSI) aims to transform interaction experience into persistent improvements, yet external feedback is costly and world-model predictions may become unreliable as policies evolve. We introduce RSI-Anything, a domain-general framework built on JEPA-Anything that uses predictive world models to guide action and learning-data selection. Observed feedback drives policy or workflow improvement and conditional world-model adaptation, while component and compatibility checks govern the retention of updated policy-model pairs. Across diverse computational and scientific domains, RSI-Anything demonstrates substantial gains. In code repair, with the coder fixed, success increases from 26.3% with a static verifier to 29.2% through predictive selection and 31.2% after world-model adaptation, at comparable test usage. Graphical user interface (GUI) evaluations showed higher task completion, while average actions decrease by 39.4%. Across five successive policy versions, laboratory-planning scores more than double from 21.80% to 46.01%. In reaction optimization, RSI-Anything achieves 96.15% mean best observed yield versus 82.16% for Bayesian optimization under the same 40-experiment budget. An end-to-end immune-science application further connects workflow evolution and predictive modeling with therapeutic hypothesis generation. These results demonstrate how predictive world models and retained system updates enable recursive improvement across diverse environments.