HDATF
日本語
相談する
メニュー

論文アーカイブ

収録論文と要約を番号の大きい順に掲載します。

収録論文 1,430資料更新日 Research Monitoringベンチマーク
  1. 832ResearcharXiv:2608.04850

    Two-dimensional Toda--Arnoldi correspondence: Holomorphic Krylov geometry and counterdiabatic transport

    • The paper proves that the diagonal and subdiagonal coefficients in the Arnoldi reduction of a non-Hermitian Hamiltonian exactly match the Flaschka variables of the two-dimensional Toda lattice.
    • This Toda structure closes without determining the remaining upper Hessenberg terms of the Arnoldi matrix.
    • The square of the subdiagonal coefficient determines the Fubini-Study metric and Berry curvature of the Krylov subspace.
    • Along the actual path, the Arnoldi frame generates exact isospectral transport and realizes counterdiabatic driving.
    • The paper confirms that the Hermitian limit reduces to the classical Toda-Lanczos correspondence.
    概要PDF
  2. 831AI / GeneralarXiv:2608.05102

    ABSeeker: Training Long-Horizon Search Agents via Answer-Backtracked Credit Assignment

    • The paper proposes the ABC framework, which backtracks from the answer to reconstruct intermediate clues and evaluates each search step against those clues.
    • The key idea is precise credit assignment: useful steps in failed trajectories are rewarded, while unnecessary steps in successful trajectories are penalized.
    • ABC-SFT uses step-wise loss weighting, and ABC-GRPO uses step-wise rewards to improve both conventional SFT and GRPO.
    • With a 4B model, it achieves state-of-the-art performance at the same scale, improving from 37.3 to 55.3 percent on BrowseComp and from 39.1 to 52.9 percent on BrowseComp-ZH.
    • Trained on only 8.5K examples, it also generalizes well to other benchmarks such as xbench and GAIA.
    概要PDF
  3. 830Computer VisionarXiv:2608.02980

    Qwen-3D: A Generalist 3D Vision-Language Model for Spatial Understanding

    • It compresses visual tokens spatially with 3D position information, including depth and camera pose, to process long videos efficiently.
    • With 3D Rotary Positional Embedding, attention operates directly in 3D space, which improves cross-view reasoning.
    • A structure where the language model and the segmentation decoder share the full visual token set greatly improves 3D visual grounding.
    • It outperforms existing 3D LMMs by 4% in 3D visual grounding at Acc@25 and by 13% in 3D instance segmentation mAP.
    • Its 2D vision-language performance stays at the level of Qwen2.5-VL, which shows that 3D training does not hurt 2D capability.
    概要PDF
  4. 829AI / GeneralarXiv:2608.03972

    ReflectRL: Learning from Golden Negative Trajectories via Reflective-to-Direct Reasoning

    • We prove that expert-model failure trajectories, GNT, are far more effective for iterative learning than failures from itself or from weaker models.
    • GNT includes both a valid prefix close to the correct answer and local error points, making it high-quality material for reflective learning.
    • Reflective-to-Direct Policy Transition gradually shifts capability from hint dependence to solving on its own.
    • It achieves consistent accuracy gains across nine benchmarks, four models from 1.5B to 8B, and four training methods.
    • It cuts response length by more than half while improving accuracy and preventing entropy collapse.
    概要PDF
  5. 828Computer VisionarXiv:2608.05070

    HelloWorld: Enabling Socially Interactive Characters in Video World Models

    • This is the first video world model to realize social interaction between a character and a user, such as waving, greeting, and speaking to the user.
    • The paper proposes a self-distillation pipeline that reuses interaction videos generated by the model itself as training data.
    • Without additional training, it controls the timing of interactions at the frame level using only cross-attention masks.
    • It builds the first social-interaction benchmark, HelloWorldBench, with 400 samples and 3 interaction metrics.
    • It substantially outperforms existing world models on interaction quality while maintaining SOTA-level video aesthetics and camera following.
    概要PDF
  6. 827Computer VisionarXiv:2608.04436

    ToolArtist: Tool-Using Unified Multimodal Models for Agentic Image Generation

    • It is the first fully agentic image generation model that unifies reasoning, external tools such as text and image search, and image generation under one UMM policy.
    • It develops an optical transformation method for SFT that hides the image-generation tool and leaves only the final image, forcing the UMM to draw directly.
    • It uses RAD-GRPO, which combines intent reward and quality reward, to reinforce the full reasoning-search-drawing trajectory.
    • With only 7,132 high-quality SFT trajectories, it outperforms partial agentic methods on WISE and WorldGenBench-Humanities.
    • It openly releases the training data and the full post-processing infrastructure.
    概要PDF
  7. 826AI / GeneralarXiv:2608.05144

    Argus: A General-Purpose Agentic Runtime for Long-Horizon Reasoning

    • It proposes a verification-gated runtime self-evolution structure that approves goal and method changes only through a verification gate.
    • The Manager, Planner, Engineer, and Reviewer roles each carry out bounded missions on durable project state.
    • On SWE-Bench Pro, it reaches 78% compared with 59% for Direct Copilot, while using 1.41x as many tokens.
    • At the mature stage, it uses 21% fewer solve tokens and shortens workflow time by 15% compared with the new Wave stage.
    • It is validated on seven GPT-5.5 benchmarks and produces practical results, including an upstream merge of the RWKV6 kernel.
    概要PDF
  8. 825Machine LearningarXiv:2608.04001

    Test-Time Scaling in Reasoning LLMs: Inference Regimes, Evaluation, and Reproducibility

    • It formalizes test-time scaling into three structural regimes: single-path, leaf-level, and prefix-level.
    • It emphasizes that the target of evaluation is the entire inference system, including prompts, decoders, verifiers, and stopping rules, not just the model weights.
    • It shows the risk of a poor selection criterion, with mean log probability accuracy dropping from 75.6% to 65.8% as the sample count increases.
    • It releases more than 2 billion inference traces, covering mathematics, science, and competitions, to provide a foundation for reproducible research.
    • It organizes the open-weight reasoning-model ecosystem by training method and interface control criteria.
    概要PDF
  9. 824Computer VisionarXiv:2608.05131

    OPD-V: Visual On-Policy Self-Distillation with Modality Balance

    • It quantifies modality imbalance in multimodal models and proposes a new paradigm that uses this text bias as privileged information.
    • It defines a modality-balanced trust region from the logit difference between a zoom-in image as the positive teacher and a masked image as the negative teacher, and it uses this region for selective distillation.
    • On Qwen3.5-4B, it improves the average score from 64.3% to 80.0%, which also beats the 397B model at 77.4%.
    • It shortens response length by 74.5% while improving accuracy, which reduces training time per sample by 25% to 32%.
    • It shows consistent gains across six benchmarks, four backbones, and five post-processing methods.
    概要PDF
  10. 823RoboticsarXiv:2608.05042

    BridgeVLA++: A Data-Efficient, Generalizable, and Memory-Augmented Vision-Language-Action Framework for 3D Manipulation

    • It is the first framework to integrate temporal memory, which remembers task order, and spatial memory, which remembers the location of occluded objects, into a 3D VLA model.
    • It maximizes data efficiency by converting point clouds into multi-view images for the VLM and predicting actions through heatmaps.
    • On memory-dependent benchmarks such as RMBench and MemoryBench, it achieves an average jump from 20 percent to 93.3 percent compared with the previous state of the art.
    • On real robots such as Franka and Dobot, it shows strong generalization with fewer than 10 training demonstrations and extends to dual-arm manipulation.
    • The heatmap intermediate representation, 2D pretraining, and input alignment are the main contributors to performance, while 3D position input actually hurts performance.
    概要PDF
  11. 822RoboticsarXiv:2608.04996

    DreamWAM: Beyond RGB Future Prediction for World Action Models

    • It extends future prediction beyond RGB to motion, depth, and semantic information, which substantially improves the robustness of robot action models.
    • The extra information is used only during training, while inference uses RGB alone, so deployment cost does not increase.
    • On standard LIBERO, performance improves slightly from 97.3 percent to 98.4 percent, while on unseen shifts in LIBERO-Plus it improves substantially from 51.4 percent to 63.4 percent.
    • Real-robot experiments show the same pattern, with standard settings improving from 90.8 percent to 96.7 percent and perturbed settings improving from 55.6 percent to 74.4 percent.
    • Among the three auxiliary signals, motion matters most, and depth and semantics work best when injected as residual information.
    概要PDF
  12. 821Computer VisionarXiv:2608.05000

    Towards Physics of Multimodal Pretraining: Knowledge Flow, Modality Synergy, Early Unification, and Recipes

    • It finds a strong asymmetry in which language and visual understanding act as priors that drive visual generation, while the reverse transfer is weak.
    • When data complexity is low, modalities are synergistic, but when complexity is high they compete, and a shared attention structure promotes synergy.
    • Early joint training is far more effective than late alignment, and late fusion causes a vision-laziness effect.
    • Using these principles, the authors create a training recipe that uses only 5 percent of the compute to train a 13.5B MoE model on 2T tokens and still achieves strong generation quality.
    • The synergy pattern remains consistent regardless of visual tokenizer design, which shows that architecture choice is the main driver.
    概要PDF
  13. 820Computer VisionarXiv:2608.02711

    Hunyuan3D-Buffalo 1.0: A Unified Multimodal Model for Scalable 3D Generation, Understanding, and Editing

    • It unifies 3D understanding, text-to-3D generation, instruction-based 3D editing, and text-based part generation in one architecture.
    • It builds one of the largest multimodal 3D training corpora, with 87 million examples in total, including 25 million for understanding, 50 million for generation, and 12 million for editing.
    • The Nano3D-v2 agent-based pipeline mass-produces geometrically consistent 3D editing data.
    • The combination of Hunyuan3D-VLM for understanding and DiT for generation achieves state of the art on both generation and editing benchmarks.
    • Experiments show cross-task synergy, where better generation and understanding improve editing quality.
    概要PDF
  14. 819Computer VisionarXiv:2608.02713

    Quo Vadis, World Modeling?

    • It expands the paradigm of world models from predicting physical state to transforming information into agent-usable information.
    • It classifies world proxies into six functional forms: dynamics, spatial, execution, memory, skill, and reward.
    • It systematizes capability enhancement in three stages: inference-time support, training-time optimization, and co-evolution between the agent and the proxy.
    • It analyzes how world proxies address the cost, risk, and limited parallelism of direct environment interaction.
    • It is a roadmap paper from a joint effort by Shanghai AI Lab, Zhejiang University, and NUS that sets the direction for world-model research.
    概要PDF
  15. 818AI / GeneralarXiv:2608.01735

    DAPD: Dual-Anchored Policy Distillation

    • It diagnoses the root cause of privilege illusion as information asymmetry between the teacher and the student.
    • It uses the self-conditioned distribution as a bridge to align reference and rollout behaviors along two information-matched paths.
    • It supervises both directions, from reference to rollout and from rollout to reference, to reduce dependence on privileged information while preserving accuracy.
    • It reduces wrong claims by 45% compared with OPSD while consistently improving reasoning performance.
    • The improvement persists as the model scales from 1.7B to 32B.
    概要PDF
  16. 817RoboticsarXiv:2608.03727

    Track4Action: Distilling World-Centric 3D Tracker into Vision-Language-Action Policies

    • It proposes a privileged-supervision approach in which a 3D tracker is used only during training and removed at deployment.
    • Track queries match the 3D tracker features extracted from Track4World and inject them into the VLA action head.
    • It reaches 82.3 percent zero-shot on LIBERO-Plus, improving by 7.6 percentage points over the unaligned version and by 3.0 points over LaMP.
    • It records 80.44 percent on RoboTwin 2.0 clean and 81.48 percent on randomized.
    • On four real dual-arm robot tasks, it reaches 67.5 percent, which is 25.0 percentage points higher than the unaligned version.
    概要PDF
  17. 816Computer VisionarXiv:2608.04010

    ParVL: Parallel Scaling and Expandable Compute Allocation for Multimodal LLMs

    • It proposes the first framework that creates separate parallel branches for ViT and LLM so that compute allocation can be adjusted independently.
    • A prefix-conditioned branch shares backbone parameters, so compute can be expanded without increasing parameters.
    • Systematic experiments on nine vision-language branch configurations show that the optimal allocation differs by task.
    • It confirms consistent gains over a single-branch model at 1B, 2B, and 8B scale.
    • It analyzes that parallel execution does not directly translate extra compute into wall-clock latency.
    概要PDF
  18. 815AI / GeneralarXiv:2608.02276

    Harness-R1: Learning to Edit Executable Runtime Harnesses from Agent Failure Trajectories

    • It proposes the first learning-based method that automatically generates executable patches for runtime harnesses from agent failure trajectories.
    • A 9B harness engineer is trained only with GRPO while the target agent weights remain frozen, which creates a stable pipeline.
    • It improves success rate by 9.3 percentage points on average across WebShop, ALFWorld, and DBBench.
    • Even after target-agent fine-tuning, it adds another 5.0 percentage points, which shows that the harness engineer and the agent can co-evolve.
    • Transfer experiments on 20 different target settings show an average improvement of 7.06 percentage points, which confirms generalization.
    概要PDF
  19. 814RoboticsarXiv:2608.03701

    LiLa-WAM: Lightweight Latent Reasoning World-Action Model for Robotic Manipulation

    • It trains a lightweight world-action model that unifies future-state prediction and action generation in a single stream on one 24 GB GPU.
    • It proposes the Visual Transition Token, or VTT, which specifies a task only through visual feature direction without language.
    • On 50 RoboTwin 2.0 tasks, it achieves a 90.48 percent success rate and shows strong efficiency relative to parameter count.
    • It also performs competitively with much larger models in LIBERO and real-robot experiments.
    • Using only a DINOV3 visual backbone, it shows that effective robot control is possible without dependence on a language model.
    概要PDF
  20. 813Computer VisionarXiv:2608.03979

    Video-DeepResearch: Towards the Next-Generation Multimodal Deepresearch Agent

    • We designed a deep-research agent pipeline that selects key scenes from video frames and performs visual search.
    • We found and fixed modality bias, where the model prefers text search only, and knowledge leakage, where it answers from memory without tools.
    • After SFT, we used GRPO, Group Relative Policy Optimization, to build autonomous exploration ability beyond imitation learning.
    • We newly built VideoDR-Bench, with 200 complex multi-hop VQA problems.
    • The 35B-A3B model reaches SOTA at 64.0%, beating Claude-4.5-Sonnet at 59.0% by five points.
    概要PDF
  21. 812Machine LearningarXiv:2608.02352

    Qwen-CUA: Native Computer Use for (almost) Everything

    • We trained a native computer-use agent that controls the keyboard and mouse from screenshots only, using a 397B-A17B MoE.
    • It keeps up to 20 screenshots and groups older ones into blocks to improve KV-cache efficiency.
    • We ran large-scale reinforcement learning with about 100,000 vCPUs, 40,000 verified tasks, and tens of thousands of concurrent environments.
    • It consistently outperformed Qwen3.7 across eight benchmarks, with 86.2% on OSWorld-Verified.
    • It also improved safety by reducing attack success rate from 36.6% to 16.4%.
    概要PDF
  22. 811Information RetrievalarXiv:2608.02738

    Knowledge-Geometry Decoupling: Refreshable Pretrained Transfer for Streaming Recommendation

    • [Withdrawn paper] The latest arXiv version, v2, states that Zixuan Wang has withdrawn this paper.
    • The official version history dates the withdrawal to 6 August 2026; the stated reason concerns industrial co-authors’ data sharing and privacy regulations.
    • Official reason: Withdrawn due to data sharing and privacy regulations of industrial co-authors.
    • Previously described methods and performance figures were pre-withdrawal claims and are not presented as currently valid research findings or grounds for recommending adoption.
    • Official withdrawal notice and version history: https://arxiv.org/abs/2608.02738 .
    概要PDF
  23. 810RoboticsarXiv:2608.03924

    ETA: A New Agentic Paradigm for Embodied Tasks

    • It proposes an agent paradigm for robots in a Planner-Tool-World loop, where the robot acts once, observes, and then decides again.
    • OpenETA provides interchangeable planners, composable tools, auditable memory, and replayable trajectories in open source.
    • Without a VLA, gpt-5.6-Sol alone solves 117 of 130 LIBERO tasks, or 90%, at PASS@5.
    • It includes a limited self-evolution mechanism that improves skills by learning from failed trajectories.
    • It demonstrates real hardware integration with a UR5e robot arm, showing applicability beyond simulation.
    概要PDF
  24. 809Computer VisionarXiv:2608.03974

    JoyAI-Video-Edit: Real-Time Open-Ended Video Editing with Autoregressive Diffusion

    • It achieves real-time 720p video editing at 30 FPS on a single B200 GPU with a 16B-parameter diffusion model.
    • SA-DMD, or source-anchoring distillation, reduces the diffusion process to two stages while preserving the fidelity of the original video.
    • Long-horizon autoregressive distillation suppresses accumulated color drift and appearance distortion in long videos.
    • It builds a new long-video editing benchmark called LongV2VBench for systematic evaluation.
    • It substantially outperforms existing streaming editors and produces results that are competitive with offline systems.
    概要PDF
  25. 808Machine LearningarXiv:2608.02585

    GradCuit: Credit-Assigned Gradient Flow Enables Robust and Interpretable Test-Time Latent Reasoning

    • It inserted the latent states to be optimized between prompt representations and generated-token representations at a selected transformer layer, while freezing the base model's parameters.
    • Through causal self-attention, reward-weighted gradients from all generated tokens flow directly to earlier latent states without a decoding detour.
    • Average accuracy across five base models, three reasoning benchmarks, and two answer formats was 64.5%, exceeding the strongest comparison method by 2.4 points and chain-of-thought prompting by 6.6 points.
    • It reduced accuracy variation across seven learning rates from 1.53 to 0.82, and gradient analysis showed that latent influence concentrates on reasoning connectives, also providing interpretability.
    • The effect was strongest in early-to-middle layers and disappeared in later layers. Evaluation was limited to three reasoning tasks, so layer selection and generalization to other tasks require verification.
    概要PDF
  26. 807AI / GeneralarXiv:2608.01418

    Reusing Rollouts under Policy Lag: Prefix-Normalized Policy Optimization for LLM Reinforcement Learning

    • PNPO multiplies likelihood ratios through each token position and then takes their geometric mean, retaining causal prefix information while narrowing the weight range.
    • Unlike GSPO, which assigns one weight to an entire response, it uses a different prefix average at each position, but it is a biased approximation that gives up exact state-action distribution correction.
    • Under fourfold reuse, its peak average across three mathematical benchmarks was 50.24, 3.00 points above GSPO, but the difference was inconsistent under single-use training.
    • Under the same budget of 2400 updates, using only 150 fresh rollout batches produced results similar to using 600 batches for single-use training, demonstrating the potential of rollout reuse.
    • The study used only one model, three mathematical tasks, and one run per setting, without separately ablating components, so the effect of prefix normalization alone and its effects on asynchronous and replay data have not yet been isolated.
    概要PDF
  27. 806Computer VisionarXiv:2608.02437

    InfiniSplat: Implicit Gaussian Decoding for Large-Baseline Monocular View Synthesis

    • It built an implicit decoder that places Gaussian support points on local surfaces estimated from monocular depth and predicts Gaussian attributes from image features queried at those positions.
    • Decoupling support-point placement and attribute prediction from a fixed pixel grid produced a structure that reduces surface cracks and scattered points even under large viewpoint shifts.
    • After training only on the synthetic indoor dataset Hypersim, it outperformed existing single-image feed-forward methods on ETH3D, ScanNet++, Tanks and Temples, and DL3DV.
    • A single forward pass produces a 3D representation that supports real-time rendering, showing potential for spatial-photo exploration and augmented- and virtual-reality displays.
    • Regions unseen in a single image remain unknown, and incorrect depth estimates propagate directly to support-point placement, leaving structural distortions on reflective, transparent, or thin objects and under extreme viewpoint shifts.
    概要PDF
  28. 805LLMs / NLParXiv:2608.02101

    Cross-Domain Hybrid OPD for Generalizable Search Agents

    • It used two-stage training: first learning autonomous planning and iterative search through agent reinforcement learning, then restoring general capabilities through on-policy distillation from experts in multiple domains.
    • Search capabilities were largely preserved after distillation, and several search tasks improved further over the reinforcement-learning stage, demonstrating a balance between specialization and generality.
    • Replacing distillation with reinforcement learning on mixed search and general data reduced HYEval3.1 logical reasoning from 46.90% to 36.31% and BBEH Mini from 47.86% to 35.22%.
    • Domain-specific teachers and a curriculum beginning with easy problems produced better results on difficult reasoning, coding, and mathematical tasks than a single generalist teacher or training without a curriculum.
    • The technical report's evidence focuses on Hunyuan3-based Yuanbao and the presented search and general benchmarks, so generalization to other base models and external operational environments requires additional validation.
    概要PDF
  29. 804ResearcharXiv:2608.02023

    SwanTale: Unified Multi-Speaker Speech and Audio Generation for Instruct and Zero-Shot Tasks

    • Speech preprocessing, multilayer captions, quality refinement, and synthesis of rare situations produced approximately 70 million training records describing environments, speakers, and content.
    • It combined SwanVAE and a flow-based transformer with quality conditioning, Engram, and a unified mixture of experts to handle reference speech and natural-language instructions in one model.
    • In single-speaker evaluation with reference speech, it achieved timbre similarity of 0.95 and content error of 0.086, improving over the earlier SwanVoice.
    • In instruction evaluations, it ranked 1st in Chinese with APS 86.1 and tied for 1st in English with APS 84.2, and could generate speech, ambient sound, and sound effects in one waveform.
    • Weaknesses remain in emotional transitions in complex background music, multi-speaker scenes longer than 2 minutes, continuous control of a particular speaker's emotion, stress, and pauses, and role acting.
    概要PDF
  30. 803AI / GeneralarXiv:2608.02163

    From Simple QA to Deep Research: A Verifiable Benchmark Constructed through Iterative Task Evolution

    • The benchmark consists of 500 deep-research tasks spanning 31 topics and 10 broad categories, and it separates different capabilities with three query types.
    • The construction loop works as follows: the Explorer solves tasks in a real environment, the Formalizer organizes the results into a DAG and scoring checkpoints, and the Challenger creates the next-round query using clues that have not yet been used.
    • As the rounds progress, the average DAG depth rises quickly from 2.5 in round 1 to 5.2 in round 5 and then levels off at about 7.0, while the number of nodes grows from 4.6 to 25.2, checkpoints from 13.6 to 123.2, and Explorer tool calls from 3.9 to 23.3. The average evolution rounds is 14.19.
    • The difficulty genuinely increases. For the two query types other than the one that gives ample information, the average round score falls from about 0.93 and 0.95 to about 0.29 and 0.40. Even so, the relative ranking of models is preserved in most rounds.
    • Removing tools lowers the score for all ten models. The overall score drops by an average of 0.14 and the evidence-based and analytical checkpoints drop by an average of 0.07 each, which means the tasks require long-tail information that is not well covered in pretraining.
    • The scoring is stable as well. On 100 manually checked tasks, repeated scoring by the same model or by different models both show high Pearson correlation with human evaluation, between 0.884 and 0.916.
    概要PDF
  31. 802RoboticsarXiv:2608.02326

    ChainVLA: Chaining Vision-Language-Action Queries through a Unified Execution State for Long-Horizon Manipulation

    • It divides execution state into two parts. The backward-looking part is progress, combining a recurrent task state and sparsely logged event memory. The forward-looking part is the unexecuted tail of the most recent prediction. This state is used only as revisable prior information, not as a fixed plan.
    • On RMBench, the 1.2B model scores 62.8 on average. Among methods that cover all tasks, the strongest Mem-0 uses 8B plus an extra 2B and gets 52.8.
    • The largest gain is on Put Back Block. This task requires putting back a block whose original location is no longer visible, and it scores 96, compared with 90 for Mem-0 and 50 for MemoryVLA.
    • The ablation results are asymmetric. Keeping only progress gives 11.2, keeping only the action tail gives 3.0, and removing both gives 1.6, while the full model gets 62.8. A half-state is not very different from no state.
    • A control that mimicked prior methods was insufficient. A fixed-length observation history combined with temporal ensembling, which smooths overlapping predictions from the back, reached 35.6, 27.2 points below the full model.
    • It shows that smoothness alone is not the cause. The linear extrapolation control has lower error than the full model on all four boundary metrics for Put Back Block and on three of four for Swap Blocks. Yet its success rates are 0% and 2%, while the full model reaches 96% and 74%.
    概要PDF
  32. 801Computer VisionarXiv:2608.02583

    UEmbed: Unified Sparse and Dense Multimodal Embeddings

    • It attaches 16 special tokens and partitions the vocabulary into 16 k-means clusters so that each token owns one chunk, and the number of special tokens N was swept across 2, 4, 8, 16, and 32, with stable performance up to 16 and a clear drop at 32.
    • It compresses the vocabulary first, merging tokens that become identical after accent removal, lowercasing, and whitespace normalization, which reduces the vocabulary size from 248,320 to 184,016, and scoring uses the maximum weight among merged tokens.
    • UEmbed-9B scores 71.8 dense and 71.0 sparse on MMEB-v2, making it the best dense retrieval model among those trained on public data and a new state-of-the-art for sparse retrieval.
    • For UEmbed-2B, using a combined score that sums dense and sparse results raises text from 53.6 to 53.9 and visual documents from 77.0 to 77.5, while natural images and videos see little benefit because they contain less surface vocabulary information.
    • On the agentic retrieval benchmark BrowseComp-Plus, sparse mode reduces the number of search rounds. UEmbed-2B drops from 39.19 dense rounds to 32.67 sparse rounds while recall rises from 53.47 to 57.04, and the 9B model keeps the same 49.76 accuracy while reducing rounds from 33.68 to 31.05.
    • The authors also state the limits honestly: modern LLMs with large vocabularies can activate odd tokens such as "_alt", and because the training corpus is biased toward English and Chinese, tokens from other languages are much less likely to appear.
    概要PDF
  33. 800Computer VisionarXiv:2608.01964

    LongHorizon-Harness: Advancing Long-Horizon Agents for Real-World Tasks

    • It identifies three main reasons for failure on long tasks: goals drift as errors accumulate, important information becomes hard to find as the record grows longer, and the system loses track of what it has done so far.
    • Its core structure is a Manage-Execute-Audit loop. The manager cannot access the environment directly and sees only state and audit reports; the executor does one subtask at a time in a fresh context; and the auditor checks the environment directly with read-only tools.
    • The only memory passed between rounds is the audit report, while the executor's raw trace is discarded. Termination happens when the audited state satisfies the original goal, when there are no more subtasks to do, when user input is needed, or when the round budget is exhausted.
    • A thin AgentAdapter swaps in a different backend without changing the original agent loop. It can mix harnesses such as Claude Code, Codex CLI, OpenClaw, and Hermes Agent, and it can mix models such as Claude Opus, GPT, and Qwen across the three roles.
    • On the combination of Qwen 3.7-Plus and Claude Code, the pass rate on WeaveBench rises from 51.8 percent to 80.7 percent, which is almost twice the 41.2 percent of the official best-reported Claude Opus 4.7 combination.
    • Across the full OSWorld 2.0 benchmark, the binary completion rate rises from 2.8 percent to 8.3 percent, about a threefold gain, and on 34 subsets it rises from 20.0 percent to 34.3 percent relative to Claude Opus 4.7.
    • In the case study, the baseline method gets 0.00 because it directly edits document XML and only makes the output look right, while the proposed method follows the specified GUI procedure and gets 0.89 after the auditor reparses the XML and checks 15 title styles.
    概要PDF
  34. 799RoboticsarXiv:2608.02547

    Why Does Action Chunking Improve Behavioral Cloning Performance in Robotic Control?

    • It tested the three existing hypotheses with controlled experiments and concluded that neither temporal consistency, reduced effective horizon, nor representation learning alone can fully explain the performance of action chunking.
    • A lag policy that predicts only one step ahead from past observations replaces much of chunking's ability to use the past and reduce accumulated error. The validation error was lowest for lags between 5 and 15.
    • The chunking policy learns relationships between actions at different time gaps from a single observation, so it behaves like an ensemble of several lagged policies. The authors call this an implicit ensemble.
    • In every simulated and real-robot environment they tested, deploying the chunking policy as a random lag ensemble matched the performance of action chunking without using action chunks.
    • An explicit ensemble built from multiple independently trained policies performed even better than chunking. They report about a 30% improvement on the Robomimic Transport task.
    • At control frequencies between 15 and 20 Hz, the lag policy worked, but it failed when raised to 50 to 60 Hz. Treating partial chunks of length 5 as a single action made it work again, and the authors explain this by noting that human vision-based behavior updates at 2 to 10 Hz.
    概要PDF
  35. 798LLMs / NLParXiv:2608.02046

    CompanionBench: A Theory-Anchored, Real-World-Grounded Benchmark for AI Emotional Companionship

    • The capability system measures 10 capabilities extracted from 25 theory papers on a 1-point to 10-point scale, with 42 sub-items in total, and it adds a reverse-layer list of 18 negative patterns with three severity levels.
    • It adds four capabilities that prior work did not score explicitly: tolerating ambiguity, self-object responsiveness, positive resonance, and appropriately calibrated challenge.
    • The second axis is self-disclosure gate replay, which measures how often disclosure is obtained and how deep it goes, and it also counts disclosures that were elicited without being originally given.
    • It is highly reproducible, with Spearman correlations of 0.996 in Chinese and 0.953 in English for the ranking of the 10 capability scores under independent reruns, and 8 of the 10 capabilities are at or above 0.93; the lowest are positive resonance at 0.829 and calibrated challenge at 0.863.
    • The gate behavior really follows the agent's behavior, because the rate of obtained disclosures and the rate of early self-disclosure are strongly negatively correlated, with Pearson correlations of -0.938 in English and -0.979 in Chinese.
    • Across the 28-model leaderboard, gpt-5.5 ranks first in English with an IRT score of 8.27, while the lowest, doubao-char-251128, scores 4.28; three models specialized for roleplay sit near the bottom, and the biggest deficits are in tolerating ambiguity, calibrated challenge, and acknowledgment.
    • The numbers show that warmth and substance are different dimensions, because claude-haiku-4-5 drops from 0.780 before penalty rules to 0.548 after penalties are applied, and the lowest model drops from 0.395 to 0.017.
    概要PDF
  36. 797Computer VisionarXiv:2608.02603

    WorldExam: Benchmarking World Models from Apparent Appearance to Inherent Reactivity

    • The benchmark spans three manipulation styles, four diagnosis stages, eight evaluation tasks, 1,474 test cases, and 20 evaluated models.
    • The eight tasks are Camera Control, Subject Control, Scene Revisit, Terrain Interaction, Object Interaction, Social Interaction, Physical Reaction, and Goal Completion. The first three test instruction following and spatial consistency. The next four test whether the world reacts on its own after seeing the scene, and the final Goal Completion task tests whether the model can plan its own execution steps from only a high-level goal.
    • It breaks the same behavior into atomic control units and converts them to the input format each model expects. Camera-based models receive SE(3) camera trajectories, action-based models receive discrete action sequences, and language-based models receive natural-language prompts.
    • In World Reactivity cases, the expected reaction is removed from the instruction. For example, the model is told only to move forward toward a staircase, and it must infer the vertical adaptation needed to climb the stairs from the scene itself.
    • To match model capabilities, the benchmark is split into two tracks. The static-scene track, where only the camera moves, evaluates all three modalities, while the dynamic-interaction track, where the subject and environment must interact, evaluates only the action and language modalities.
    • From the Table 1 comparison, existing benchmarks have many cases but narrow task coverage. WorldScore has 3,000 cases and one task, iWorld-Bench has 4,900 cases and two tasks, while WorldExam covers all eight tasks with only 1,474 cases.
    • The conclusion is that high image quality and good instruction following do not imply world reactivity. No model combined broad task coverage with consistently high performance.
    概要PDF
  37. 796AI / GeneralarXiv:2608.02287

    SKT: Skill-Use Training at Scale via Verified Synthetic Data Generation

    • From 2,000 public skills, it created 4,000 task packages and 27,164 verified execution trajectories, and it also built the SkillEval benchmark from a separate skill pool that does not overlap with the training data using the same pipeline.
    • As the skill budget used for training increases from 0 to 100, 500, 1000, and 2000, the Qwen3.5-9B SkillEval score rises monotonically to 55.24, 59.8, 67.4, 70.8, and 72.48.
    • The benefit transfers across execution environments: a model trained on OpenCode trajectories improves from 4.94 to 9.29 on SkillsBench in DeepAgents, and training on same-environment trajectories raises it to 13.62, meaning it retains 50.1 percent of that gain.
    • A single checkpoint trained on mixed trajectories from two environments beats the original models in all 8 paired comparisons between benchmarks and execution environments, with gains ranging from 2.64 to 18.81 points. SkillEval rises from 55.24 to 74.05 on OpenCode and from 51.62 to 69.96 on DeepAgents.
    • The mixed-trained model stays within an average of 2.71 points of the environment-specific model and even surpasses it in 3 of the 8 comparisons, which means separate models per environment are less necessary.
    • When the 77 SkillsBench tasks are grouped by the number of required skills, one-skill tasks improve from 9.54 to 16.94, two-skill tasks from 3.95 to 20.72, and tasks with three or more skills from 4.04 to 12.08, so the gain is not limited to single-skill tasks.
    概要PDF
  38. 795RoboticsarXiv:2608.02304

    TRACE: Ergodic Trajectory Optimization for Active Scene Reconstruction

    • On all eight Replica scenes, it achieves higher PSNR than ActiveGS, with an average gain of 1.5 dB. The image-quality metrics LPIPS and SSIM move in the same direction.
    • The gains differ by scene type. In office scenes, SSIM is similar but LPIPS improves by 6% to 19%, while in room and hotel scenes the SSIM gap widens from 0.012 to 0.017.
    • The gain does not come simply from taking more observation frames. A variant that added 10 random shots on the ActiveGS path matched the original within 0.08 dB, while a variant with 10 evenly interpolated shots was actually 1.7 dB worse.
    • If the footstep-depletion trick is removed and only the original kernel ergodic objective is used, performance drops by 2.1 dB and becomes worse than ActiveGS on 6 of the 8 scenes. This shows that suppressing redundant observations is the key.
    • On a Unitree Go2 quadruped robot with an Intel D435 RGB-D camera, the system was run on a 42-square-meter space for a 5-minute mission, and it executed without infeasible commands at the planned horizons.
    • The real-world execution success rate was 100%, while the compared NBV baselines lightly touched the environment in every attempt. Because the trajectory itself is the optimization variable, waypoints can be passed directly to the controller.
    概要PDF
  39. 794AI / GeneralarXiv:2608.02365

    Faster-WAM: Do World Action Models Need Deep Action Modules?

    • The main issue is speed. Even Fast-WAM, a representative low-latency model, takes 211.7 ms for one action chunk, about three times the 71.4 ms of π0.5, a VLA model.
    • DoT removes the one-to-one alignment between video layers and action layers required by prior MoT methods, so action-head depth can be set independently of backbone depth and can access representations from all backbone layers.
    • On the average of four LIBERO task groups, it reaches 98.5%, 0.9 points above Fast-WAM's 97.6%, and gains 2.6 points on LIBERO-Long. It also outperforms π0.5, X-VLA, and Motus, all of which use robotics pretraining.
    • Under the same 24GB consumer GPU setting, latency is 66.5 ms for Faster-WAM, 68.2 ms for π0, 71.4 ms for π0.5, 105.3 ms for X-VLA, and 211.7 ms for Fast-WAM. LingBot-VA and Motus could not be measured because they ran out of memory.
    • On out-of-distribution generalization in LIBERO-Plus, it reaches 75.0% and beats Fast-WAM's 51.5% by 23.5 points. Camera perturbation rises from 16.4% to 67.9%, sensor noise from 37.7% to 82.7%, and language perturbation from 68.9% to 92.1%.
    • Ablation starts at 49.5% and climbs step by step to 60.3% with simple DoT using only the final layer, 66.8% with KV-Fusion, 71.3% with RoPE alignment, and 75.0% after removing text cross-attention. Layer mixing signals are concentrated in middle layers but spread widely across all layers.
    概要PDF
  40. 793RoboticsarXiv:2608.02580

    Ego2Robot: Scalable Robot Data Synthesis from Egocentric Human Data

    • The pipeline creates virtual fingertips from 21 hand keypoints using a weighted index and middle finger combination of 0.7 and 0.3, extracts the gripper center and opening width, and removes temporal noise with Savitzky-Golay filtering and SLERP. For videos without hand-pose annotations, it estimates pose with WiLoR and DynHaMR, and Qwen3.5 splits long videos into sub-tasks.
    • From about 1,940 hours of raw video, it produces 18,561 hours of data for 15 robot morphologies, which is about a 9.6x expansion.
    • In pretraining mixture experiments, the 1:1 mix ranks first in five of seven settings, reaching RoboTwin Clean 68.1% plus 5.9, Randomized 53.5% plus 2.6, and task-semantic axis 54.1% plus 7.9, while the 1:3 mix brings almost no gain.
    • By factor, lighting adds 7.6 points, robot color adds 6.4, and camera offset adds 5.9, which means visual robustness gives the largest gain. Unseen objects rise from 29.3% to 40.0% with the 3:1 mix, paraphrasing reaches 68.5%, and Franka stays below 7%, which shows transfer to robots with very different kinematics failed.
    • In the ablation that uses only egocentric data and no robot data, raw egocentric video scores 28.1%, the pipeline raises it to 31.7%, increasing the number of morphologies from 1 to 15 raises it to 33.5%, and adding the raw egocentric data raises it further to 37.3%.
    • On five long-horizon tasks with a real ARX ACone robot, the best result came from converting 20 teleoperation demonstrations into about 7-minute egocentric play videos per scene and mixing them, which improved block insertion by 14 points and screw insertion by 13 points.
    概要PDF
  41. 792ResearcharXiv:2607.29660

    A Simple Necessary and Sufficient Condition for Yang--Baxter Integrability

    • It gives a rigorous proof, for translation-invariant nearest-neighbor spin chains, of a conjecture that had been open for 40 years: that the Reshetikhin condition is necessary and sufficient for Yang-Baxter integrability.
    • The condition can be checked directly at the Hamiltonian level and is equivalent to conservation of the total energy flow, so it is also experimentally observable.
    • It establishes a quantum version of the Liouville-Arnold theorem, showing that Yang-Baxter integrability is equivalent to the existence of infinitely many local conservation laws.
    • When searching for new integrable models, one can check the Hamiltonian directly instead of finding an R-matrix, which greatly simplifies the search.
    概要PDF
  42. 791RoboticsarXiv:2607.29302

    BWM: A Low-Cost High-Fidelity World Simulator for Robot Learning

    • An action-conditioned world model predicts future scenes conditioned on robot actions, combining initial environment guidance, dynamic history, and precise action control.
    • We built a training-data pipeline that preserves action-observation alignment through trajectory replay, duplicate clip sampling, and enhanced initial observations.
    • It ranks first overall among open-source entries in WorldArena Challenge Track 1 plus two Track 2 applications.
    • The policy trained on BWM-generated data reaches a 71.00% average success rate on a real robot, a large improvement over the previous best world simulator at 53.33%.
    • We release model checkpoints, training and inference code, and data-generation and policy-evaluation interfaces as open source.
    概要PDF
  43. 790Computer VisionarXiv:2607.29627

    FlexComposer: Unified Video Compositing from Images to Dynamic Footage with Flexible Trajectory Control

    • It proposes a unified video synthesis framework that handles still images and dynamic videos within the same framework.
    • It standardizes heterogeneous inputs with a unified normalized foreground representation that separates the foreground's intrinsic motion from global movement.
    • It uses spatially aware latent injection based on the translation equivariance of the VAE latent space to achieve precise trajectory control without extra trainable modules.
    • It learns lighting and shadows naturally from a hybrid dataset that combines simulation, real movie footage, and generated data, together with a synthetic-to-real curriculum.
    • It outperforms prior state-of-the-art methods in visual quality, temporal consistency, and trajectory adherence.
    概要PDF
  44. 789LLMs / NLParXiv:2607.27497

    SkillSmith: Learning to Compose Parametric Skills and Textual Knowledge

    • It treats model weights, or prefix cache, as a modality on par with text so that the LLM can read and reason over both together.
    • SkillSmith takes source-task weights, text metadata, and a target description as input, and directly synthesizes new prefix weights tailored to the target.
    • It builds a new synthetic dataset called Composite SNI to verify the effectiveness of combining text and weights on new tasks that require composing skills from two tasks.
    • With 4B Gemma 3, it beats all baselines that use only text or only weights, and on MMLU-ProX it even beats methods based on downstream fine-tuning.
    • It shows that retrieval-based methods can synthesize weights effectively in noisy environments even without exact source mapping.
    概要PDF
  45. 788RoboticsarXiv:2607.29613

    WCM: A World Critic Model for Vision-Language-Action Reinforcement Learning

    • It defines the critic's fundamental limitation in VLA reinforcement learning, when it sees only a single frame, as the state approximation problem, and shows that scalar value regression cannot learn temporal structure.
    • WCM is built on a LeJEPA structure and is explicitly trained so that the critic's representations capture temporal dynamics by predicting future latent states and estimating value at the same time.
    • It is compatible with modern VLA backbones such as π0, π0.5, and OpenVLA-OFT, and can be integrated into both on-policy and off-policy training.
    • It achieves consistently best performance on 149 tasks across four benchmarks, with especially large gains in out-of-distribution generalization.
    • On seven real-robot manipulation tasks, it confirms stable performance with OpenVLA-OFT and π0.5-based off-policy RL.
    概要PDF
  46. 787RoboticsarXiv:2607.28993

    ST-WAM: Semantic-Temporal World Action Model for Robust Manipulation under Visual Distribution Shifts

    • The authors discover a hallucination phenomenon in which world action models hallucinate training-scene content in visually unfamiliar environments, and they confirm that 70.6 percent of 180 inspected cases exhibit this behavior.
    • They show through 290-frame triplet experiments that DINOv3 features are more robust to visual changes than VAE latent representations while still separating task states well.
    • A dual-space future expert predicts both VAE and DINO spaces, and a current-anchor intention retrieval module finds task-relevant evidence from the recent DINO history.
    • It achieves 98.7 percent on LIBERO and 92.8 percent on RoboTwin 2.0, and on visually perturbed environments such as LIBERO-Plus it improves by 21.3 points over Fast-WAM and raises real-robot success from 25.8 percent to 61.5 percent.
    • It is trained end to end without extra pretraining or task-specific annotation, and it is efficient because it does not require explicit future generation at inference time.
    概要PDF
  47. 786Computer VisionarXiv:2607.29679

    Scaling Properties of Text Conditioning in Visual Generation

    • The paper proves that natural-language prompts stop improving image quality as they get longer because they do not add more information, whereas structured prompts continue to add information and improve quality linearly.
    • It introduces two information measures, GPG for image-text alignment and ED for attribute precision, and uses them to quantify the relationship with diffusion loss mathematically.
    • It trains an LLM prompter with SFT, cold start, and verifier-based reinforcement learning so that short user requests are automatically converted into information-rich structured prompts.
    • The final system outperforms every open-source model evaluated on compositional, reasoning, and world-knowledge benchmarks, and it matches or exceeds closed top-tier models.
    • Fine-grained editing, such as object moves, material changes, and scene replacement, becomes possible through field edits in the structured prompt.
    概要PDF
  48. 785AI / GeneralarXiv:2607.28527

    MANTA: Multi-Agent Network Topology Adaptation for Self-Evolving Multi-Agent Systems

    • We propose structural self-improvement, which changes the communication topology in real time while solving the task instead of fixing it before execution.
    • It reuses structures learned from past experience before a task starts and partially revises them when problems arise during execution.
    • Across five benchmarks, it averages 74.0 points, 5.8 points higher than the previous best.
    • Rather than simply increasing the number of agents, it improves structure by rewiring connections or adding verification steps.
    • Meta-level operations, meaning structure planning and revision, account for only about 12% of total token usage, so it is efficient.
    • It also has the lowest token usage among multi-agent systems.
    概要PDF
  49. 784LLMs / NLParXiv:2607.28229

    EMBL AI Librarian: Life-Sciences Knowledge Layer for AI Agents

    • It builds a natural-language knowledge layer for AI agents on top of Europe PMC, which contains 40 million records.
    • A single LLM orchestrates the full pipeline, from query planning to execution, paper parsing, and evidence-snippet extraction.
    • It handles the process through live search instead of a dense vector database that would require 744 GB, which greatly reduces infrastructure cost.
    • On ScholarQA-Bench, it improves Citation F1 by 16 points, and on LitQA2 it raises accuracy from 70.3 with GPT-5.4 plus web search to 78.9 with GPT-5.4 plus the Librarian.
    • Applied to the ProClaim verification pipeline, it improves agreement with expert consensus by 5 points, and the code is public.
    概要PDF
  50. 783Machine LearningarXiv:2607.28022

    Flux-OPD: On-Policy Distillation with Evolving Contexts

    • It proposes Flux-OPD, which uses evolving contexts as supervision signals during training in open-ended domains.
    • Through reverse KL decomposition, it shows that the student is distilled by a geometric mean teacher and defines a conflict term.
    • The context correction strategy does not use an unstable context teacher as the direct target, and instead injects only the difference signal into a stable reference teacher.
    • It uses the conflict term as a weighting signal, correcting strongly when contexts agree and weakly when they conflict.
    • It achieves 19.63 to 20.61 on HealthBench and 81.48 to 82.26 on VBench for 1.3B, with consistent gains across three independent runs.
    概要PDF