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. 532Information RetrievalarXiv:2606.31693

    ShopX: A Foundation Model for Intent-to-Item Fulfillment in Agentic Shopping

    • It unifies intent-to-item fulfillment in a single model: intent understanding, execution planning, and SID-native control.
    • Semantic ID (SID) design: semantically recoverable and editable by the LLM.
    • It specializes a general LLM into a shopping agent through continued pretraining and post-training recipes.
    • On single-turn and multi-turn evaluations based on Taobao production logs, it outperforms existing tool-mediated systems.
    • It achieves especially large gains on complex and ambiguous requests, reaching precision of 88% to 97%.
    • The model-native design reduces loss during handoff between orchestration and item execution.
    概要PDF
  2. 531AI / GeneralarXiv:2606.29537

    OSWorld2.0: Benchmarking Computer Use Agents on Long-Horizon Real-World Tasks

    • It contains 108 long-horizon workflows, with a human median of 1.6 hours and an average of 318 tool calls per agent.
    • Claude Opus 4.8 with max thinking completes 20.6 percent of tasks, while GPT-5.5 completes about 13 percent, although its token efficiency is better.
    • It defines 10 challenge phenomena, including dynamic environments, cross-source reasoning, hidden-state reasoning, and conflict resolution.
    • It builds a realistic environment from 31 self-hosted web services, including email, banking, and chat.
    • The main failure causes are forgetting constraints, missing intermediate information, guessing, and skipping verification, not basic GUI manipulation.
    • Agents spend less than 7 percent of their own budget on error detection and correction.
    概要PDF
  3. 530Machine LearningarXiv:2606.32026

    AdaJEPA: An Adaptive Latent World Model

    • The first test-time adaptation framework for a JEPA world model, using a Plan-Execute-Adapt-Replan loop.
    • It performs self-supervised adaptation from observed transitions alone, without additional expert demonstrations.
    • It achieves large gains with only one gradient step per MPC replanning step.
    • It produces significant improvements over the fixed model on both in-distribution and out-of-distribution target-reaching tasks.
    • It shows especially large gains when training data is limited.
    • It compares online buffer management strategies using recent-N and hard-N approaches.
    概要PDF
  4. 529AI / GeneralarXiv:2606.31410

    Xiaomi-GUI-0 Technical Report

    • A hybrid, real-device-centered infrastructure uses physical devices as the main environment and a sandbox as a secondary one to match the real deployment distribution.
    • An error-driven data flywheel turns failed rollout trajectories into corrective actions, reflection explanations, and recovery demonstrations.
    • A three-stage progressive curriculum: SFT for basic operations, Step RL for long-term planning, and Agentic RL for error recovery.
    • It achieves a 72.0% success rate on RealMobile and 78.9% on AndroidWorld.
    • It strengthens the ability to handle abnormal states in real applications, including account states, permission dialogs, payment authentication, and risk controls.
    • It builds a closed-loop real-device environment to solve emulator-detection mechanisms.
    概要PDF
  5. 528LLMs / NLParXiv:2606.30406

    MOPD: Multi-Teacher On-Policy Distillation for Capability Integration in LLM Post-Training

    • It uses a three-stage pipeline: General SFT, Domain-specialised RL, and Multi-teacher On-Policy Distillation.
    • It aligns teacher and student on student rollouts with per-token reverse KL, which gives dense supervision and remains on-policy.
    • On Qwen3-30B-A3B, it improves by 5.5 points over the next best baseline and closes 91% to 95% of the teacher-student gap.
    • Domain teacher training can run in parallel, which increases development throughput and decouples the domains.
    • Teachers from the same origin are key to stable optimization, while using external teachers risks divergence.
    • It is deployed in the MiMo-V2-Flash industrial model, which validates its utility at frontier scale.
    概要PDF
  6. 527AI / GeneralarXiv:2606.28446

    Domain-Informed Multi-View Self-Distillation for Astronomical Light-Curve Representation Learning with JEPA

    • C-RoPE, or Continuous Rotary Positional Embedding, uses rotary position encoding to handle irregular observation intervals.
    • EANE, or Error-Aware Numeric Embedding, integrates measurement uncertainty or noise into the embedding.
    • It performs multi-view self-distillation across three domain views: raw light curves, GLS periodograms, and phase-folded curves.
    • On the StarEmbed benchmark, it beats hand-crafted features on 15 of 16 metrics, with macro-F1 ranging from 42.56 to 63.58.
    • It is validated on downstream applications such as similarity search, zero-point drift detection, and parameter estimation.
    • It reaches state of the art on 5 of the 12 non-astronomy datasets in PYRREGULAR, which highlights the importance of domain-specific design.
    概要PDF
  7. 526AI / GeneralarXiv:2606.30626

    DOPD: Dual On-policy Distillation

    • It defines the privilege illusion as a failure pattern in which information asymmetry from privileged information is mistaken for ability gaps.
    • Advantage-aware routing dynamically selects the supervision source, teacher or student, for each token.
    • It improves average performance by +7.5 points over Vanilla OPD on LLMs and +6.0 points on VLMs.
    • It applies strong-teacher distillation over the full vocabulary and weak-student self-supervision by token category.
    • It shows consistent gains in continual learning, OOD generalization, and training stability.
    • Token-level analysis shows that high-teacher, low-student tokens are responsible for transferring core abilities.
    概要PDF
  8. 525RoboticsarXiv:2606.28276

    SimFoundry: Modular and Automated Scene Generation for Policy Learning and Evaluation

    • It automatically generates a sim-ready digital twin from a single video, with F1 scores from 0.81 to 0.92 and 0.93 to 0.99 after 3 minutes of tuning.
    • It generates scene variations that preserve affordances while increasing diversity across object, scene, and task cousins.
    • Across 7 tasks and 5 policies, simulation evaluation correlates with real performance at Pearson 0.911, which is 0.59 higher than the previous SOTA.
    • Training policies on cousin data improves zero-shot sim-to-real transfer success by 17 to 40 percent.
    • It shows strong zero-shot transfer on YAM, with 99 percent Pot on Stove success, and on DROID, with 100 percent Stack Dishware success.
    • It handles more complex tasks than prior work, including multi-step manipulation, articulated-object interaction, and bimanual collaboration.
    概要PDF
  9. 524Information RetrievalarXiv:2606.29718

    Diagnosing and Mitigating Context Rot in Long-horizon Search

    • Context rot is defined as the phenomenon in which accumulated context causes the model to give up or produce an uncertain answer.
    • It performs a systematic diagnosis using four open-source models, including Qwen3.5-397B and GLM-4.7, across three benchmarks.
    • It evaluates seven management methods, including context compaction, trimming, and isolation, in terms of performance, cost, and their effect on rot.
    • Rot-aware filtering, which removes uncertain or give-up trajectories, improves average performance by 2.6 to 4.9 percent.
    • The combination of compaction and trimming offers the best balance between cost and rot mitigation.
    • Context isolation through subagents is highly model dependent and only works well on stronger models.
    概要PDF
  10. 523Machine LearningarXiv:2606.29723

    ScaleAware-JEPA: Latent Representation for Discovery in Multiscale Physical Fields

    • Constrained Diffusion Decomposition, or CDD, decomposes fields into pixel-level scale components.
    • The CDD scale determines the masking geometry, aligning the physical scale hierarchy with the prediction task.
    • It is applied to three heterogeneous domains: MHD turbulence, the NGC 3627 molecular gas field, and Chengdu night lights.
    • It can generate a dense latent atlas without labels, segmentation, or catalogs.
    • It quantitatively verifies the superiority of scale-aware masking over fixed patch masks.
    • It is an independent study performed on consumer hardware, specifically an Apple M3 Pro and an RTX 3090.
    概要PDF
  11. 522Machine LearningarXiv:2606.28277

    Towards Automating Scientific Review with Google's Paper Assistant Tool

    • It uses a four-stage agent pipeline of Segmenter, Adaptive Budgeting, Deep Review, and Global Synthesis.
    • It improves recall by 34 percent over zero-shot on math error detection in the SPOT benchmark.
    • It was piloted as a pre-submission tool at two major conferences, STOC and ICML.
    • It proposes a four-level taxonomy for AI roles: Tool, Supporting Reviewer, and Fully Automated.
    • It addresses the precision drop and context limits of Pass@k through segment-level deep review.
    • A NeurIPS 2021 study found that even human reviewers had a 23 percent disagreement rate, which suggests practical value for AI review.
    概要PDF
  12. 521LLMs / NLParXiv:2606.30616

    Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent

    • A 35B MoE shows a performance path based on expanding agent horizon rather than parameter count.
    • Knowledge-Action Graph (KAG): a long-horizon trajectory infrastructure that stores evidence, actions, observations, and verification results as connected objects.
    • A three-stage training pipeline: all-domain SFT, domain-specific teacher RL, and domain-routed multi-teacher on-policy distillation.
    • It generates agent trajectories averaging 45K tokens and integrates six heterogeneous domains.
    • It outperforms 1T models on SEAL-0 (56.4), IFBench (80.6), and FrontierScience-Olympiad (79.0).
    • Salient Vocabulary Alignment reduces conflicts between reasoning patterns across domains.
    概要PDF
  13. 520Computer VisionarXiv:2606.30534

    Orca: The World is in Your Mind

    • It proposes Next-State Prediction as a unified paradigm instead of Next-Token, Next-Frame, or Next-Action Prediction.
    • It designs two complementary paradigms, unconscious learning through dense state transitions in continuous video and conscious learning through language-conditioned sparse event transitions.
    • It builds a large-scale world-learning dataset with 125,000 hours of video and 160 million event annotations.
    • It freezes the encoder and trains only a lightweight decoder to produce text, image, and action readouts.
    • It validates scalability at the 4B and 0.8B scales, showing that larger world latents translate directly into stronger downstream performance.
    • It also openly states clear limitations, including the current restriction to vision and language modalities, short-horizon transitions, and limited model scale.
    概要PDF
  14. 519RoboticsarXiv:2606.27374

    World Action Models Enable Continual Imitation Learning with Recurrent Generative Replays

    • The first continual imitation learning approach to use WAM generative capability, reconstructed future frames, as pseudo-replay.
    • Compared with sequential fine-tuning, it reduces catastrophic forgetting by up to 50% and comes close to experience replay that uses real demonstrations.
    • Recursive generation feeds future observations predicted by the WAM back into the model to synthesize complete trajectories, which is recurrent generative replay.
    • It remains effective on real robots as well: NBT improves from 96.3 to 60.5, reducing forgetting by 40%, and FWT improves from 50 to 80.
    • It identifies two main bottlenecks: visual quality degradation during long-horizon recursive generation, reflected in lower PSNR, and mismatch between imagined and grounded actions, 83% versus 42%.
    • Action representation drift shows that REGEN, at 0.12, preserves representations much better than Seq-FT, at 0.30.
    概要PDF
  15. 518RoboticsarXiv:2606.28133

    Translation as a Bridging Action: Transferring Manipulation Skills from Humans to Robots

    • The paper argues that rotation information from human hands is noisy and that rotation-inclusive 6DoF learning is suboptimal because the human hand and parallel gripper have different contact patterns.
    • It adopts relative wrist translation in the head-camera frame as a bridging action and builds a shared action space between humans and robots.
    • An interleaved action token sequence, 3D-wrist to 6D-eef to gripper, together with attention masking, handles missing components in heterogeneous data.
    • The training strategy has three stages: large-scale human-data pretraining, joint human-robot training, and a small amount of real-robot fine-tuning.
    • Adding bridging-action supervision to robot data is essential, raising overall success from 12.5 percent to 38.3 percent.
    • The upper-bound analysis shows that removing the embodiment gap lets the bridging objective achieve even stronger transfer performance, with overall progress rising from 59.75 percent to 73.54 percent.
    概要PDF
  16. 517RoboticsarXiv:2606.27581

    SceneBot: Contact-Prompted General Humanoid Whole Body Tracking with Scene-Interaction

    • Using reference motion plus per-link contact labels as policy conditioning, we unify free-space and contact-rich behaviors in a single policy for the first time.
    • Hindsight scene reconstruction infers the scene-interaction graph from retargeted human motion and reconstructs terrain and object assets in reverse.
    • With 7.5 hours of contact-rich training data, combining AMASS, OMOMO, Bones, and Lafan, it generalizes to unseen motions and environments.
    • Global root tracking is essential: without it, terrain success drops from 100% to 45% and object-task success from 95% to 20%.
    • Removing contact labels makes object-grasp success converge to 0%, proving the decisive role of contact labels.
    • On the real Unitree G1, state-estimation performance is confirmed with an average root-position error of 0.032 m and a rotation error of 0.018 rad.
    概要PDF
  17. 516AI / GeneralarXiv:2606.26859

    AgentX: Towards Agent-Driven Self-Iteration of Industrial Recommender Systems

    • The four-stage closed loop is Brainstorm, which proposes evidence-based ideas; Developing, which generates repo-grounded code; Evaluation, which performs guardrail-vetoed A/B decisions; and Harness Evolution, which applies SGPO.
    • SGPO, or Semantic-Gradient-based Prompt Optimization, uses execution trajectories to update agent prompts and doubles weekly throughput.
    • Three AgentX workers turned 374 ideas into 10 launchable rollouts over 3 weeks, which is 8x the concurrency of human engineers and 3.7x the business value.
    • It uses online A/B feedback as a reward signal so that the work stays aligned with real business value, including a 0.561 percent increase in app dwell time and more than 100 million yuan in annual value.
    • It shows that the approach can extend beyond recommendation strategy to model-side research, such as paper replication, module pruning, and architecture combination.
    • It includes a negative-result assetization mechanism that turns failed experiments into structured knowledge assets.
    概要PDF
  18. 515Computer VisionarXiv:2606.28128

    PhysisForcing: Physics Reinforced World Simulator for Robotic Manipulation

    • We experimentally identify two sources of physical errors: deformation of moving objects and unnatural spatiotemporal correlations between interacting entities.
    • We design complementary pixel-level trajectory alignment loss based on point tracking and semantic-level relation alignment loss based on a frozen video encoder.
    • We propose a physics region focus strategy that concentrates physical supervision on interaction-critical regions, outperforming uniform application by 1.5 points.
    • On R-Bench, we improve Wan2.2-I2V-A14B by 22.3% and Cosmos3-Nano by 9.2%, corresponding to gains of 7.1% and 3.7% over vanilla finetuning, respectively.
    • In the WorldArena action-planner protocol, closed-loop success rises from 16.0% to 24.0%, and average downstream policy success rises from 68.2% to 72.8%.
    • PF-Cosmos achieves the best overall score on R-Bench, PAI-Bench, and EZS-Bench.
    概要PDF
  19. 514Computer VisionarXiv:2606.27608

    Qwen-Image-2.0-RL Technical Report

    • It fine-tunes a VLM with pointwise scoring and chain-of-thought reasoning to build task-specific composite reward models for T2I, including alignment, aesthetics, and portrait quality, and for editing, including instruction following and face ID.
    • It introduces a hybrid CFG strategy in the GRPO-based RL framework to preserve pretrained knowledge and training stability at the same time.
    • On-policy distillation, or OPD, integrates the T2I teacher and the editing teacher into a single student model through trajectory-level velocity matching.
    • OPD achieves better performance than mixed RL, which optimizes multiple rewards at once, because it avoids cross-task interference.
    • It reports consistent improvements on Qwen-Image-Bench with 57.84 points, up 2.61, on the T2I arena with an Elo gain of 78, and on the editing arena with an Elo gain of 93.
    概要PDF
  20. 513RoboticsarXiv:2606.26025

    In-Context World Modeling for Robotic Control

    • We formulate the problem of restoring missing system configuration psi, such as camera viewpoint and morphology, as context in the VLA setting.
    • We prepend N task-independent random-search clips, typically 3 to 5, as context to implicitly infer system dynamics without parameter updates.
    • Across six OOD viewpoints, the average success rate is 25.0%, a gain of 5.2 percentage points over Multi-View's 19.8%, and wrong context can reduce performance to 18.9%, proving that the model truly depends on context.
    • With 20 to 80 mm spacers attached to the robot gripper, a morphology change is preserved with a 60% margin, with ICWM at 14.4% versus 5.6% for Multi-View.
    • Reusing context hidden states with KV caching restores inference latency to the 0.112 s baseline.
    • This context adaptation ability does not emerge naturally in standard imitation learning and requires an explicit training incentive.
    概要PDF
  21. 512Machine LearningarXiv:2606.26294

    The Red Queen Gödel Machine: Co-Evolving Agents and Their Evaluators

    • Controlled utility evolution, with a fixed criterion within each epoch and replacement only at epoch boundaries, preserves self-improvement guarantees even under nonstationary utility.
    • On coding tasks such as Polyglot, adding agent-as-a-judge code review signals improves pass rate by 1.8 percentage points over the previous SOTA while using 1.35x to 1.72x fewer tokens.
    • In the paper-writing setting, the co-evolved writer's acceptance rate rises from 21.8 percent to 40.5 percent, which is a 1.86x improvement.
    • In IMO-level proof grading, the co-evolved grader reaches higher accuracy with three times less search cost than the fixed baseline.
    • By injecting adversarial goals at epoch boundaries, it corrects reviewer self-preference bias toward AI-generated papers, where AI papers are accepted 1.91 times too often.
    • As the evaluator gets stronger, it also creates a curriculum effect for the task agent, which leads to archive reranking and gradual strengthening.
    概要PDF
  22. 511AI / GeneralarXiv:2606.26300

    The Verification Horizon: No Silver Bullet for Coding Agent Rewards

    • When quality judging and trajectory monitoring are added to unit-test rewards, the SWE-Bench hacking solve rate drops sharply from 28.57% to 0.56%, while the normal solve rate rises from 40.22% to 60.53%.
    • For frontend tasks, introducing a browser-based interaction evaluator defends against length attacks that are vulnerable to static code inspection.
    • User-feedback rewards achieve up to a 13.3-point improvement across five internal coding-agent benchmarks.
    • For autonomous-evaluator data filtered for long-horizon tasks, RFT gains +1.91 points over random sampling, from 21.61 to 23.52.
    • There is no single mechanism that satisfies all three dimensions of verification, scalability, fidelity, and robustness, so task-specific design is essential.
    • Verification is not auxiliary but core training infrastructure, and it must co-evolve continuously with policy capability growth.
    概要PDF
  23. 510AI / GeneralarXiv:2606.27226

    Ask, Don't Judge: Binary Questions for Interpretable LLM Evaluation and Self-Improvement

    • It uses a meta-prompt structure that breaks evaluation criteria into atomic binary questions, which greatly improves transparency and diagnostic value.
    • On QAGS for factual consistency, it improves Spearman rho by 0.195 over UniEval, which shows that the decomposition itself contributes most of the information gain.
    • BINEVAL catches subtle factual errors that G-Eval and UniEval give a perfect 5.0 to by asking 7 questions each, reaching 1.57 points and coming close to human judgment at 2.0.
    • It avoids the usual ceiling effect of LLM judges and distinguishes borderline outputs from clearly defective ones more effectively.
    • It improves both summarization and IFBench prompts through cross-model prompt updates that use question-level feedback.
    • Its task-agnostic and training-free design makes it immediately applicable to summarization, dialogue, instruction following, and other tasks.
    概要PDF
  24. 509Machine LearningarXiv:2606.25086

    Training for the Model You Return: Improving Optimization for Iterate-Averaged Language Models

    • Formulate iterate-average optimization as a continuous-time stochastic quadratic optimal control problem.
    • PACE is a lightweight wrapper on top of AdamW, with extra memory equal to one copy of the model weights.
    • The standard SCO convergence rate is guaranteed in the convex setting, and arbitrarily large bounded errors can be improved in the quadratic setting.
    • It shows consistent gains in three fine-tuning settings on SmolLM2, Qwen3, and Gemma3, and also works for GPT-2 FineWeb pretraining.
    • It is competitive with Schedule-Free and removes the need for LR decay, outperforming WSD across all token budgets with constant LR.
    • Extensive ablations on pullback strength c, EMA power κ, and update frequency show its robustness.
    概要PDF
  25. 508RoboticsarXiv:2606.27295

    LA4VLA: Learning to Act without Seeing via Language-Action Pretraining

    • It provides a systematic analysis of vision-language asymmetry in VLA training at both the data level and the input level.
    • It automatically builds 33K LA episodes from existing demonstrations without collecting extra data, in LA4-33K.
    • LA-only pretraining consistently outperforms VLA-only pretraining at the same scale.
    • Mixed LA-VLA pretraining reaches 87.53% on MetaWorld, 96.28% on LIBERO, and 83.3% on real robots.
    • Under visual noise, the LA-pretrained policy achieves a success rate that is 25 percentage points higher than VLA-only.
    • t-SNE analysis of the LA-pretrained policy shows clearly separated representations by command direction.
    概要PDF
  26. 507Machine LearningarXiv:2606.27326

    Hallucination in World Models is Predictable and Preventable

    • It defines three types of hallucination: perceptual hallucination from the tokenizer, action hallucination from marginalization in the dynamics, and scene hallucination from rollout divergence.
    • MMBench2 contains 65,600 trajectories spanning 427 hours, 210 tasks, and 10 domains, together with ground-truth actions, rewards, and the simulator.
    • Coverage-aware training improves both the tokenizer and dynamics, which raises rollout PSNR by 0.88 dB.
    • Collecting only 50 curiosity-driven trajectories yields 0.325 adaptation performance on unseen tasks, which is about 90 percent of the expert or human score of 0.362.
    • The hallucination predictors show a strong negative Spearman correlation of about 0.80 with rollout ΔPSNR.
    • The analysis also compares an off-the-shelf tokenizer, Wan 2.1 VAE, with an in-domain tokenizer.
    概要PDF
  27. 506Machine LearningarXiv:2606.26217

    Fast LeWorldModel

    • Action-prefix prediction generates multi-horizon latents in parallel, removing the bottleneck of sequential rollouts.
    • The dynamics module time drops from 31.4 seconds to 8.0 seconds, a 3.9x speedup, and overall CEM solve time falls by 48%.
    • Both the initial open-loop latent loss and its growth slope are significantly lower than in LeWM.
    • Average planning success rate rises from 85.8% to 90.5%, including 98% on Two-Room, 96% on Push-T, and 80% on Cube.
    • An ablation confirms that dense prefix supervision is essential for performance, not just terminal-only prediction.
    • In physical-state probing, an MLP probe achieves the best accuracy on agent position, block position, and angle.
    概要PDF
  28. 505LLMs / NLParXiv:2606.26790

    OPID: On-Policy Skill Distillation for Agentic Reinforcement Learning

    • Because it directly extracts hindsight skills from completed on-policy trajectories, no external skill library or retrieval is needed.
    • It introduces a two-level hierarchy with episode-level skills for global workflows and failure-avoidance rules, and step-level skills for local decisions.
    • Critical-first routing uses step-level skills at important moments and falls back to episode-level skills otherwise.
    • It achieves 58.9 percent on ALFWorld, compared with 46.1 percent for GRPO, and 74.2 percent on WebShop, compared with 63.3 percent for GRPO.
    • It provides a qualitative analysis showing that skill injection fixes the hallucinated-target error in GRPO-trained agents, where the agent searches for objects that do not exist.
    • It is a training-time-only framework that needs no analyzer, external skill retrieval, or privileged context at inference time.
    概要PDF
  29. 504AI / GeneralarXiv:2606.24937

    The Hitchhiker's Guide to Agentic AI: From Foundations to Systems

    • It covers the foundations from LLM architecture, including tokenization, transformers, attention, and GPU systems, all the way to Flash Attention.
    • It covers alignment techniques such as RLHF, PPO, DPO, and GRPO, and reasoning techniques such as Chain-of-Thought and test-time scaling.
    • For agents, it covers trajectory RL, RAG and Agentic RAG, in-context, external, episodic, and semantic memory systems, and agent harness design.
    • It systematically organizes MCP, or Model Context Protocol, A2A, or Agent-to-Agent, protocols, along with tool use and agent skills.
    • It covers centralized, distributed, and hierarchical multi-agent architectures and communication patterns among agents.
    • It covers the full practical stack, from evaluation methods and agent UI design to production deployment, across more than 600 pages and 400 references.
    概要PDF
  30. 503Computer VisionarXiv:2606.26907

    Qwen-Image-Agent: Bridging the Context Gap in Real-World Image Generation

    • It defines the root cause of real-world T2I failures as the Context Gap and proposes an agent paradigm to address it.
    • It structures Context-Aware Planning into three stages, information-level, content-level, and generation-level, for identifying missing context, collecting it, and allocating it.
    • Context Grounding collects context from four channels: reasoning, web and image search, memory, and feedback.
    • It builds IA-Bench with four capabilities, Plan, Reason, Search, and Memory, 17 tasks, 730 test instances, and 1,801 checklist items.
    • The intelligence of the MLLM backbone is decisive for the whole system, and swapping the backbone causes large drops on most metrics.
    • It finds that overusing image search can hurt generation quality, which shows that the search boundary must match the generator's ability.
    概要PDF
  31. 502Machine LearningarXiv:2606.25971

    Improving Neural Network Training by Decoupling the Magnitude and Direction of Weight Vectors

    • It mathematically analyzes interference between weight magnitude and direction: direction updates are inversely proportional to the current magnitude, and magnitude grows as a byproduct of direction learning.
    • It decomposes weights inside the optimizer into fixed-norm directions plus learnable gains, so it can be applied without changing the model architecture.
    • Weight decay and warmup become unnecessary, greatly simplifying large-scale training recipes.
    • The optimal learning rate transfers across model widths, so no LR retuning is needed, similar to µP but obtained directly.
    • On large MoE models, MuonMD reaches the same loss as AdamW with about half the compute.
    • Fixing the direction on the sphere captures most of the improvement, and learnable gains provide additional accuracy gains.
    概要PDF
  32. 501LLMs / NLParXiv:2606.25331

    Improved Large Language Diffusion Models

    • It is the first case of scaling a masked diffusion language model to 12T tokens, trained from scratch with 8B parameters.
    • Grouped-query attention reduces KV-cache memory and tied embeddings slim the parameters down to 7.62B.
    • Confidence-based scoring improves PIQA by +1.3, ARC-C by +0.6, and HellaSwag by +2.3 over likelihood-based scoring on multiple-choice evaluation.
    • Extending SFT to 12 epochs yields continuous gains on GSM8K, MATH, and MMLU-Pro, confirming data reuse effects in diffusion LLMs.
    • The base model reaches MMLU 74.8, BBH 71.3, and GSM8K 81.9, surpassing Qwen2.5 7B across multiple benchmarks.
    • The instruct model is competitive without RL alignment, with MMLU-Redux 76.4 and GSM8K 89.0, but it still trails Qwen2.5 Instruct on code and math.
    概要PDF
  33. 500Computer VisionarXiv:2606.27377

    DanceOPD: On-Policy Generative Field Distillation

    • The method formalizes the conflict between text-to-image, local editing, and global editing from a velocity-field perspective and resolves it with on-policy distillation.
    • It removes target-field ambiguity with hard routing that assigns exactly one capability field to each sample.
    • On GEditBench, it improves by 8.1 percent over the existing OPD baseline while preserving GenEval performance comparable to the T2I source.
    • For local-plus-global editing combinations, it improves by 16.1 percent over the best competing baseline and by 7.9 percent over the local-edit source.
    • When absorbing the realism field, it improves by 9.9 percent over off-policy distillation and reaches 85.3 percent of the student-teacher reward gap.
    • It also shows that classifier-free guidance can be absorbed under the same velocity-MSE objective.
    概要PDF
  34. 499LLMs / NLParXiv:2606.24775

    Are We Ready For An Agent-Native Memory System?

    • We propose an analytical framework that systematically decomposes agent memory into four modules: representation and storage, extraction, retrieval and routing, and maintenance.
    • We evaluate 12 memory systems plus two baselines on a unified testbed spanning five workloads and 11 datasets.
    • No single architecture is universally best. Hybrid systems are strong for conversational QA and graph-based systems are strong for factual recall, but they are weak at temporal reasoning.
    • Memory extraction should preserve context at write time. Broader retention, rather than overly selective extraction, is better for downstream reasoning.
    • Conservative merging outperforms delayed flushing or excessive summarization in long-term stability.
    • Highly structured systems pay tens to hundreds of times more in index-build time and query latency, but the accuracy gains do not scale proportionally.
    概要PDF
  36. 496Machine LearningarXiv:2606.24143

    AsyncOPD: How Stale Can On-Policy Distillation Be?

    • AsyncOPD systematically studies the stale-data problem in asynchronous On-Policy Distillation, or OPD.
    • It shows that the KL direction changes the stale-data problem, where teacher-weighted forward KL is robust and student-weighted reverse KL is fragile, and it shows that recomputing advantages at training time is the most effective fix for reverse-KL OPD.
    概要PDF
  38. 494Computer VisionarXiv:2606.24253

    TuringViT: Making SOTA Vision Transformers Accessible to All

    • TuringViT introduces Turing linear attention (TLA) to efficiently process high-resolution and dynamically resized visual sequences.
    • Its VISTA-Curation pipeline curates image and video data to improve data efficiency and support native dynamic-resolution training.
    概要PDF
  39. 493Machine LearningarXiv:2606.24020

    You Don't Need to Run Every Eval

    • The authors build a public score matrix over 84 frontier models and 133 benchmarks, with 2,604 cells and 23.3% coverage, and show that it is effectively rank-2.
    • They design BENCHPRESS, a rank-2 ALS matrix completion method in logit space, to recover hidden scores within 4.6 points.
    概要PDF
  40. 492Computer VisionarXiv:2606.25041

    Wan-Streamer v0.1: End-to-end Real-time Interactive Foundation Models

    • Wan-Streamer models language, audio, and video as both inputs and outputs with a single transformer, and uses block-causal attention to support incremental streaming.
    • It jointly learns cognition, reasoning, generation, response timing, turn management, and cross-modal synchronization in one model instead of relying on separate VAD, ASR, language, TTS, avatar, and video generation modules.
    概要PDF
  41. 491Computer VisionarXiv:2606.24805

    DDStereo: Efficient Dual Decoder Transformers for Stereo 3D Road Anomaly Detection

    • This is the first stereo 3D detector to achieve real-time inference at 23.5 ms, at the level of monocular methods.
    • The dual decoder separates foreground binary detection using disparity features from 3D attribute regression using left-view features.
    • Shared object queries align the two decoders and produce anomaly scores without NMS or anchor matching.
    • On the KITTI-AR-OoD open-set benchmark, it outperforms S3AD with 78.09% OoD AP3D on the R40 metric.
    • It achieves SOTA compared with all existing stereo methods on the pedestrian and cyclist classes.
    • Using only an additional 10% of synthetic data, it still reaches 54.65% OoD AP3D, far exceeding S3AD's 24.45% and showing strong generalization.
    概要PDF
  42. 490LLMs / NLParXiv:2606.24530

    NatureBench: Can Coding Agents Match the Published SOTA of Nature-Family Papers?

    • It covers 90 carefully selected tasks from 5,500 Nature-family papers across six scientific domains: cell omics, proteins, biological modeling, physical modeling, molecular design, and relational reasoning.
    • NatureGym is an automatic pipeline that builds a containerized task environment directly from a paper, which ensures reproducibility.
    • Even the strongest agent, Claude Opus 4.7, exceeds the published SOTA only 17.8 percent of the time at g greater than 0.1 and matches it 47.8 percent of the time, which shows that the benchmark relies more on translation than discovery.
    • In 45.5 percent of successful paths, the key step was methodological translation rather than scientific invention, meaning conversion into familiar supervised-learning form.
    • The main failure causes are wrong method selection at 45.1 percent and insufficient compute budget at 24.4 percent, not lack of task understanding.
    • An information firewall hides the original authors' methodology, which forces independent discovery instead of reproduction.
    概要PDF
  43. 489LLMs / NLParXiv:2606.23525

    Self-Compacting Language Model Agents

    • Fixed token-threshold compression can trigger in the middle of reasoning or retrieval and accidentally delete verified facts.
    • The method combines a compression tool that the model can call directly with a lightweight rubric that says to compress when a subtask is solved and suppress compression during reasoning.
    • When the tool is provided without the rubric, usage varies widely across models and is often invoked at the wrong time, which shows that the rubric is essential.
    • On math benchmarks such as IMO-Answerbench and HMMT, it improves by up to 18.1 points over the no-compression baseline.
    • On agentic search tasks such as BrowseComp, it improves by 5 to 9 points while reducing token cost by 30 to 70 percent per question.
    • Adaptive compression is possible at inference time alone, without any fine-tuning or external supervision.
    概要PDF
  44. 488RoboticsarXiv:2606.24742

    World Value Models for Robotic Manipulation

    • This is the first approach to reuse the spatiotemporal prior knowledge of world models as the basis for robotic value estimation.
    • It combines a video-generation stream and a value DiT through a Mixture-of-Transformers, or MoT, to minimize representation interference.
    • It formulates the value function as a flow-matching-trained distributional chunk, which provides dense learning signals.
    • It releases Suboptimal-Value-Bench, which consists of 800 suboptimal trajectories plus human labels.
    • It achieves SOTA on both the expert VOC and Suboptimal-Value-Bench, and it also improves downstream policy learning consistently.
    概要PDF
  45. 487Computer VisionarXiv:2606.24888

    DiffusionBench: On Holistic Evaluation of Diffusion Transformers

    • It builds NanoGen, a unified training and evaluation framework that can switch from ImageNet to text-to-image training with only 12 lines of configuration changes.
    • After training 21 latent diffusion models, it confirms almost no correlation between ImageNet rank and text-to-image rank, with Pearson coefficients from -0.377 to -0.580.
    • It provides empirical evidence that better ImageNet FID does not translate into better text-to-image quality.
    • It releases a unified codebase that supports RAE, VAE, pixel-space, and MeanFlow.
    • DiffusionBench is a holistic benchmark that combines ImageNet and text-to-image results and is recommended as the default reporting standard for future DiT research.
    概要PDF
  46. 486AI / GeneralarXiv:2606.24855

    OpenThoughts-Agent: Data Recipes for Agentic Models

    • The authors run more than 100 controlled ablation studies on a six-stage SFT data pipeline to quantify the effect of each stage.
    • Instruction selection is the most important factor in the pipeline, which shows that the strongest model is not necessarily the best teacher.
    • Fine-tuning Qwen3-32B on 100K examples yields 54.0 percent on SWE-Bench Verified and 26.2 percent on Terminal-Bench 2.0.
    • The average score is 44.8 percent, which is 3.9 points better than the previous best open-data agent model, Nemotron-Terminal-32B at 40.9 percent.
    • It shows scaling behavior that beats competing open datasets across all training-set sizes.
    • It confirms that an 8B model trained with SFT and RL in two stages outperforms both pure SFT and pure RL.
    概要PDF
  47. 485AI / GeneralarXiv:2606.23595

    SPIRAL: Learning to Search and Aggregate

    • It is the first framework to jointly optimize sequential, parallel, and aggregation compute within a single reinforcement learning pipeline.
    • Set reinforcement learning trains parallel traces so that they are collectively useful for aggregation.
    • It achieves up to 11x scaling efficiency in pass@k versus GRPO and 15 percent better performance when parallel and aggregation scale together.
    • It points out that existing training paradigms optimize only sequential reasoning, leaving a gap with test-time scaffolds.
    • Experiments with Qwen3-4B-Instruct validate the method, and the authors identify 8B-plus parameter scaling as future work.
    概要PDF
  48. 484Computer VisionarXiv:2606.23688

    Lift4D: Harmonizing Single-View 3D Estimation for 4D Reconstruction In-the-Wild

    • We apply causal latent conditioning to a monocular 3D reconstruction model to ensure temporal consistency across frames.
    • We convert temporally consistent initialization into a deformable 3D Gaussian Splatting representation.
    • We use occlusion-aware rendering supervision, which localizes occluded regions with depth cues and handles visible and invisible regions separately.
    • We apply a view-conditioned diffusion prior to the unseen regions to complete plausible appearances.
    • On the Consistent4D synthetic benchmark, we achieve SOTA with LPIPS 0.116 and FVD 592.44.
    • On DAVIS in-the-wild videos, we demonstrate superior motion accuracy over prior methods, with CLIP 0.715 and EPE 0.161.
    概要PDF
  49. 483Computer VisionarXiv:2606.23489

    MeshFlow: Mesh Generation with Equivariant Flow Matching

    • It directly represents meshes as triangle soup, avoiding the limitations of autoregressive serialization such as slow speed and error accumulation.
    • It designs an Equivariant DiT architecture that preserves both face permutation invariance and vertex cyclic invariance.
    • It introduces a coupling loss based on nested optimal transport that removes learning signals that violate symmetry.
    • It achieves the best 1-NNA on three of the four ShapeNet categories and matches the quality of state-of-the-art autoregressive models.
    • It speeds up inference by 18.55x, enabling sub-second mesh generation compared with autoregressive methods.
    • A post-processing denoiser reduces self-intersection rate, Ri, by about 56% with negligible additional latency.
    概要PDF
  50. 482RoboticsarXiv:2606.24884

    InSight: Self-Guided Skill Acquisition via Steerable VLAs

    • It automatically splits human demonstrations into VLM planning segments and end-effector trajectories, making the VLA controllable at the primitive-action level.
    • The VLM identifies missing primitive actions in a new task and builds a flywheel that tries them through low-level control, collects success data, and retrains.
    • It achieves 92% and 96% end-to-end success rates on twisting and following without human demonstrations, compared with 32% and 16% for CaP-X.
    • It reaches 80% success on long-horizon tasks chained from 14 primitive actions, such as twisting followed by following.
    • Adding new primitive actions does not reduce existing pick-and-place skill, confirming that forgetting is not a problem.
    • In a Mars-robot scenario, it learns a sweeping behavior autonomously from only mopping demonstrations and succeeds on all 5 evaluations.
    概要PDF