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. 982Information RetrievalarXiv:2608.21156

    Graph Engineering in the Era of LLM Agents: From Individual Intelligence to System Intelligence

    • From the perspective of 'the limits of individual intelligence → system intelligence,' it presented the next step after prompt, context, harness, and loop engineering.
    • It organized graph-based design along four axes: task organization (what to do), agent coordination (who does it), runtime state management (how it runs), and system evolution.
    • It provided a framework for approaching practical challenges in long-running systems, such as failure localization and recovery and structural improvement based on execution histories, through graph-based state management.
    • A separate chapter covers scientific-discovery and lab-automation applications, making it directly useful for designing autonomous-lab agents.
    • It published lists of related papers, datasets, and projects on GitHub for use as practical references.
    概要PDF
  2. 981RoboticsarXiv:2608.21204

    Beyond Imitation: Self-Improving Robot Policies via Off-Policy Q-Planning

    • By freezing the large robot policy, with billions of parameters, and updating only a Q-function of approximately 1 billion parameters, it made self-improvement cost proportional to evaluator size rather than policy size.
    • The Q-function can learn from both successful and failed rollouts, raising real-world task performance to 90%/80% where SFT retrained only on successful data had plateaued at 55%/30%.
    • At inference time, it handles 64 candidate actions sampled by the policy with a single planning step of Q-score-weighted averaging, taking 400ms and fitting within the real-time control budget.
    • Under the same online budget, it was the only method that reliably improved from failures when compared with Best-of-N, filtered SFT, IBRL, DSRL, and DAWR.
    • It also clearly stated its limitations: it still cannot solve tasks where the base policy produces no successful candidates, and it requires an episode-level success verifier.
    概要PDF
  3. 980AI / GeneralarXiv:2608.19652

    Can Agent Memory Systems Track Evolving State?

    • It defined 'state drift': a failure in which an agent answers using old or invalidated values even when the required facts are present in retrieval results. It showed that this persists even under perfect retrieval, establishing it as a problem separate from 'retrieving well.'
    • It built StateMemBench with 234 multi-session scenarios and used closed-choice scoring to distinguish 'current state/old state/other failure,' cleanly isolating state-tracking failures.
    • StateMem explicitly tracks value replacement and dependencies (recomputation of derived values). It improved current-state accuracy 1.8-fold on DeepSeek-V4-Flash relative to the strongest same-backbone baseline (from 0.205 to 0.363), and 1.6-fold on Qwen-3.5-9B relative to the strongest existing memory system (from 0.149 to 0.233).
    • Even a single-call wrapper around existing memory systems improved all 6 backends by +32–+67 points. Comparisons with length- and cost-matched controls isolated +15–+32 points of this as the effect of the 'state structure' itself.
    • Judge labeling also confirmed that state drift accounts for a prominent share of failures on existing benchmarks such as LongMemEval.
    概要PDF
  4. 979LLMs / NLParXiv:2608.19726

    Projector Is All You Train

    • In a 3D MLLM, freezing both the encoder and language-model backbone and training only the projector, a 3-layer MLP, still matched existing baselines on 3D classification and captioning.
    • Under the same budget of 16 GPU-hours on a single A100, projector-only training performed similarly to jointly training the backbone with LoRA while delivering approximately 2× the training-sample throughput.
    • Jointly training the backbone severely degraded existing capabilities: the Llama backbone's GSM8K score fell from 86.96 to 0.61 and HumanEval from 64.02 to 0.00. Projector-only training leaves the backbone untouched, inherently avoiding this regression.
    • It analyzed apparent improvements in some vision scores after joint backbone training as artifacts of multiple-choice scoring, which renormalizes option probabilities, while instruction-following failures actually increased.
    • It proposed a modular multimodal setup as a future direction, independently attaching and training modality-specific projectors on a single language-model backbone.
    概要PDF
  5. 978LLMs / NLParXiv:2608.20169

    Task-CoEvolve: Efficient Harness Optimization via Adaptive Validation Task Selection

    • Existing harness optimization is expensive because it evaluates all validation tasks at every iteration. This work was the first to formalize the problem of evolving validation-task selection itself together with the harness.
    • Based on the observation that tasks on which candidate harnesses differ in success or failure (tasks with high success-rate variance) are most informative for distinguishing candidates, it used variance-weighted sampling to focus evaluation near the boundary of agent capabilities.
    • It corrects for sampling probabilities to estimate the full score from partial evaluation, allowing fair comparisons across iterations even when a different subset is evaluated each time.
    • On Terminal-Bench 2.1, it achieved final performance similar to full search with 20% of the evaluation budget, substantially reducing token use (for example, from 741M to 246M) and search time (from 38.0 hours to 20.5 hours).
    • At the same 20% budget, final performance was 3.3 points higher than with random sampling, confirming that 'what is evaluated' determines efficiency.
    概要PDF
  6. 977AI / GeneralarXiv:2608.20256

    Learning When to Think: Adaptive Reasoning for Test-Time Compute Allocation

    • The model directly selects one of three modes, NoThink, Short, or Long, as the first token of its answer, and this routing is trained jointly end to end within reinforcement learning (GRPO), without a separate router.
    • It combines mode-specific token limits of 1,024/3,000/unlimited with a mode-balancing reward to prevent routing collapse, where one mode crowds out the others.
    • It showed that the Short mode achieved higher accuracy than Long, demonstrating that the router correctly allocates problems according to difficulty.
    • On MATH500, it maintained accuracy of 0.782, compared with the previous 0.796, while reducing average response length by 41%, from 4,796 to 2,811 tokens.
    • Without retraining, it saved 76% of tokens on GSM8K, while maintaining equivalent accuracy without reducing the budget on the difficult AIME benchmark, showing that difficulty-based allocation worked as intended.
    概要PDF
  7. 976AI / GeneralarXiv:2608.20314

    MidTool: Mid-training Data Synthesis for Agentic Tool Use

    • It open-sourced MidTool-Mix, a 20.3-billion-token mid-training corpus dedicated to tool use that combines web content, PDFs, code, and real API/MCP skills, along with its construction pipeline.
    • Document-based 'grounding' synthesis and real-tool-execution-based 'execution' synthesis play different roles, and both were required for improvements across all eight metrics.
    • Both 4B and 8B models showed consistent gains on three benchmarks, BFCL, tau2-Bench, and MCP Universe, under either SFT or RL post-training.
    • The mid-trained 4B and 8B models outperformed official Qwen3 models on MCP-Universe, showing that dedicated mid-training can be more effective than increasing model size.
    • Execution-trajectory synthesis alone delivered its largest gain on BFCLv3 at +7.9, while grounding synthesis was stronger for transfer to unfamiliar environments such as tau2-Bench and MCP-Universe.
    概要PDF
  8. 975ResearcharXiv:2608.20328

    The Memory Hidden in Response Fluctuations: Trajectory-Level Fluctuation-Response Theory and Inequalities for Non-Markovian Jump Dynamics

    • For non-Markovian jump processes whose event rates depend on past history, it exactly decomposed each event into a predictable tendency and an unpredictable martingale increment.
    • It proved that martingale increments form a complete orthogonal basis for the fluctuations of all trajectory observables, with each coefficient matching the history-dependent response strength.
    • It enabled testing whether proposed memory coordinates sufficiently explain the response, using a nonnegative response-heterogeneity gap that remains after averaging over histories.
    • In BK ion-channel data, memory ranked by dwell-time correlations and memory ranked by responses could have opposite orderings, showing that correlations alone cannot identify important memory.
    • The experimental application relies on observational comparisons rather than interventions and on thresholded, finite-length event sequences, so it may be affected by hidden variables and missed events; explosive dynamics are outside the theory's scope.
    概要PDF
  9. 974Machine LearningarXiv:2608.17163

    Q-Learning With World Models

    • It used the world model only for tree search at action-selection time, rather than as a source of training data, gaining sample efficiency without accumulating bias from imagined trajectories.
    • Because the policy and Q-function are trained only on real-environment data, world-model errors do not enter the learning process.
    • On the Robomimic and LIBERO robot-manipulation benchmarks, it significantly outperformed strong prior methods in both sample efficiency and final performance.
    • Applying search to both training-data collection and evaluation produced the most consistent gains, and shallow search with depth 2 and a short-future weighting (lambda=0.2) was optimal.
    • In image-observation (pixel-based) settings, it also generally learned faster and achieved higher success rates than the base algorithm.
    概要PDF
  10. 973AI / GeneralarXiv:2608.20316

    Pandora's AI Model Routing Box: Efficient Allocation with Costly Value Estimation

    • It formalized the model-routing problem of deciding when to call an expensive value estimator as the Pandora's box problem from economics.
    • Under a Gaussian-signal assumption, it computed the value of information and reservation prices in closed form to build a centralized Pandora router.
    • The Pandora router used far fewer expensive estimator calls while maintaining allocation quality similar to exhaustive evaluation.
    • It validated centralized and auction-based allocation in three domains: multiple language models, retrieval-augmented experts, and mathematical models with adjustable reasoning time.
    • The approach is limited to Gaussian assumptions and two-stage estimators, and in distributed auctions, strategic experts' gains can undermine overall efficiency when evaluations of other parties are inaccurate.
    概要PDF
  11. 972AI / GeneralarXiv:2608.20318

    AI4AI-Bench: Benchmarking LLM Agents in Algorithmic Design for Recursive Self-Improvement

    • It created a benchmark that freezes 10 research repositories covering 10 training-algorithm families, lets an agent modify the training code for 4 hours, then retrains from scratch for up to 12 hours and scores the result with a hidden evaluator.
    • The average score across 29 configurations of 6 systems was 0.166, and even the best score of 0.250 closed less than one-fifth of the gap between the original algorithm (0.1) and the optimum (1.0).
    • Of 263 submissions with changes, 141 did not modify the training procedure at all. The 122 that did scored an average of 0.226, substantially outperforming the rest (0.126).
    • Increasing reasoning effort raised the proportion of attempts that addressed the algorithmic layer from 8% to 64%, and the average score from 0.094 to 0.196.
    • It released the task suite, evaluators, and all scored submissions so that evolving systems can be measured again using the same yardstick.
    概要PDF
  12. 971ResearcharXiv:2608.20066

    Resolution of Singularities in Positive Characteristic: Frobenius-Hasse Towers and Exceptional-History Descent

    • It replaces marked Rees algebras with their differential-integral saturations and constructs total Hasse operators, coefficient cubes, and filtered Rees complexes, incorporating information into the local theory that numerical order alone misses in positive characteristic.
    • It records transformations under admissible ordinary blowups through semilinear Frobenius-Hasse sources and finite exceptional genealogies, routes local defects through a six-line ledger to dedicated procedures for surfaces, toroidal-monomial, binomial, and additive cases, and then assembles their results into a single global ordering.
    • The central mechanism is a global replacement certificate that transfers owner, parent, quotient, trace, and resumption information between macroblocks; for a complete state equipped with this certificate, it claims that iteration terminates because multisets strictly decrease under a single well-ordered dependency ordering.
    • From this, it claims strong principalization of coherent ideals on smooth finite-type schemes over a perfect field k, making each ideal generated by one element, in Part IX, Theorem 15.3, and resolution preserving an ordered simple-normal-crossings boundary at every step; it states that the result is presentation-independent and functorial under re-embedding and étale pullback.
    • However, this is a single-author v1 preprint spanning nine parts and 798 pages, classified on arXiv under General Mathematics (math.GM), so whether the proof actually holds can only be judged through peer review and independent verification.
    概要PDF
  13. 970Machine LearningarXiv:2608.20061

    Let's Scale Step by Step: Compute-Efficient Hyperparameter Transfer for Large-Scale Mixture-of-Experts

    • In stage 1, it formalized μP for MoE models using MLA attention and the Muon optimizer, and showed that the optimal learning rate found on a base proxy model (0.6B total, 0.3B active) transfers unchanged to models with 2, 4, and 8 times the width (up to 30.7B total, 3.6B active).
    • In stage 2, it stops training during the stable phase without a decay phase, extracts multiple checkpoints using EMA (exponential moving average) weights, fits optimal learning rates over 255B to 502B tokens using log-log linear regression, and extrapolates to longer horizons.
    • The regression fit was R²=0.95, and it predicted 3.85×10⁻⁴ as the optimal learning rate for training on 10 trillion tokens.
    • Using the predicted learning rate, it trained a foundation model with 155B total and 17B active parameters from scratch on 10 trillion tokens. The target training run required roughly 98 times the total computation of all proxy experiments, so avoiding a search saves substantial cost.
    • However, the authors state that validation is limited to the MLA and Muon combination, batch size was deliberately left out, increasing sparsity together with width prevented isolation of the sparsity axis's effect, and an exhaustive search to verify whether the predicted learning rate is truly optimal is computationally infeasible.
    概要PDF
  14. 969ResearcharXiv:2608.20013

    Exact Fluctuation-Response Relations for Underdamped Langevin Dynamics

    • It derived exact finite-time fluctuation-response identities for underdamped Langevin dynamics that hold under arbitrary time-dependent driving and for general accumulated observables.
    • Using a Doob martingale, it exactly decomposed observable variance into a component attributable to the initial state and a component dynamically generated by subsequent noise.
    • Choosing a friction perturbation in the direction of the irreversible probability current yields an uncertainty relation saturable even at finite times and a variational principle for total entropy production.
    • In driven free diffusion, the friction-response-based factor maintained a universal lower bound even when the existing mean-current-based factor collapsed exponentially with dissipation.
    • The results are a theoretical analysis of underdamped Langevin processes and accumulated observables, so applicability to other noise structures, nonadditive observables, and actual experimental data requires separate validation.
    概要PDF
  15. 968Machine LearningarXiv:2608.19013

    Harness Continual Learning: Continual Adaptation Beyond Model Parameters

    • It formalized the HCL paradigm, which shifts the target of continual learning from model parameters to the harness (prompts, memory, tools, and routing).
    • It introduced 'harness forgetting' and controlled it through guarded evolution, which commits changes only after they pass checks for improvement, historical retention, and validity.
    • Setting the historical-loss budget to 0 yielded average forgetting of 0.39, while leaving it unlimited yielded 3.45, allowing explicit control over the stability-plasticity tradeoff.
    • It achieved relative improvements of at least 10% across 3 domains, text, multimodal, and open-world, with a highest overall multimodal score of 68.92%.
    • Component-removal experiments showed that experience memory and task-interface updates contributed most to performance and retention.
    概要PDF
  16. 967AI / GeneralarXiv:2608.19880

    EnvHarness: Awakening Static Worlds for Agent Learning

    • It proposed an EnvHarness layer that changes only behavior by wrapping an existing environment at the reset/step interface with Stage, Contract, and Chain plugins, rather than constructing a new environment.
    • By inheriting the original verifier, the ground-truth scorer, unchanged, it inherently prevents the ground-truth reliability problem of LLM-generated environments.
    • It created an automated loop in which EnvRigger diagnoses policy-failure trajectories, automatically synthesizes environments targeting those weaknesses, and validates them with new rollouts.
    • Across 5 benchmarks in 4 domains, including SWE-bench Verified, it achieved gains of up to +9.0 points and a 9.8% reduction in execution steps over learning in the original environments.
    • Through iterative policy-environment co-evolution, performance continued to rise from 47.67 to 54.79 even with 300 environments, going beyond the point where conventional environment scaling falters.
    概要PDF
  17. 966ResearcharXiv:2608.20223

    Self-Normalizing Denominators in Rational Causal Estimation

    • It defined a 'self-normalizing' class of causal-estimation denominators whose standardized values are constant across samples, and proved that such denominators cannot have weak-identification first-order limits.
    • It showed that products of powers of nested covariance minors have this property in every dimension and permit exact Wishart pivots.
    • The converse classification is complete in dimension 2, and in dimension 3 it proposed a factor-rank criterion covering instrumental-variable, front-door, and proxy formulas.
    • It proved that Wald inference for front-door adjustment remains asymptotically valid even when the residual variance of the mediating variable vanishes.
    • An audit of real medical data (right-heart catheterization) showed that 'naive diagnostics' and 'denominator-related diagnostics' lead to different conclusions.
    概要PDF
  18. 965Computer VisionarXiv:2608.20336

    WithEveryone: Unified Planning and Identity Grounding for Group Image Generation

    • It built a unified model that generates group photos of 5–10 reference individuals in one pass. Existing publicly available techniques were limited to a maximum of 4 people.
    • Layout CoT, which plans each person's position, and a location-grounded identity loss (LG-ID) prevent face matching from breaking down as the number of people increases.
    • It surpassed GPT-Image 2 in target-face similarity, 0.499 versus 0.462, and reduced copy-paste artifacts threefold, from 0.169 to 0.055.
    • With 97.3% coverage of requested people and a 2.8% duplication rate, it delivered the most balanced results among the compared systems.
    • It isolated the planning stage, rather than image generation, as the source of remaining errors, indicating the direction for further improvements.
    概要PDF
  19. 964RoboticsarXiv:2608.19613

    What Matters for Latent Actions in Robot Learning

    • It compared 41 latent action model (LAM) design choices under unified conditions across 3 simulation benchmarks + a real Franka robot.
    • A latent-action dimension of 32 was optimal for both single-arm and dual-arm robots, and additional regularization was unnecessary when pretraining regularization was done properly.
    • Fine-tuning the VLM backbone with latent actions beforehand raised success on 4 real-robot tasks from 64.75%→79.25% (+14.5 percentage points).
    • The LAM-tuned model reached 85% success after just 10k steps, surpassing the baseline's performance at 40k steps (76.25%) and demonstrating data efficiency.
    • The simple reconstruction (FDM) metric was a more reliable predictor of latent-action quality than the conventionally used MLP-probe metric.
    概要PDF
  20. 963LLMs / NLParXiv:2608.19799

    SWE-bench Science: Can Coding Agents Resolve Engineering Tasks in Science?

    • It assembled 119 tasks from 98 repositories across 20 scientific fields, including chemistry, materials science, and biology, to evaluate scientific-software repair capabilities.
    • Even the best agent achieved only a 47.9% single-attempt success rate, with a large gap between the public-test score of 96.6% and the full success rate.
    • It classified failures into four types: insufficient scientific knowledge or abstraction, superficial repairs, incomplete integration, and failure to generalize to new cases.
    • In one case, adding scientific information reduced GPT-5.6's success rate from 36.3% to 31.9%, showing that knowledge must be validated alongside execution evidence.
    • The number of tasks per field remains small, and analysis of how scientific knowledge is used in actual repairs is still preliminary, so cross-field rankings and generalization require cautious interpretation.
    概要PDF
  21. 962Computer VisionarXiv:2608.20334

    Swift-Image: Exploring the Performance Frontier of Compact Unified Image Generation Models

    • It built a unified 6B model that handles text-to-image generation, single-image editing, and multi-image editing with one set of weights.
    • When the prompt enhancer (PE) converted requests into visual task instructions, the knowledge-intensive benchmark score jumped from 2.49→4.63. The bottleneck in complex generation was intent interpretation rather than rendering.
    • Structural pruning retained nearly all performance while reducing the model to 3B, and the few-step distilled model (Turbo) actually achieved a higher overall editing score (4.20 vs 4.16).
    • Training required only approximately 243K GPU hours, showing that strong performance is possible without tens of billions of parameters.
    • It summarized practical lessons such as 'evolve the training-data distribution to match model capabilities' and 'train conflicting objectives separately before integrating them.'
    概要PDF
  22. 961Computer VisionarXiv:2608.20335

    4DAnyone: Create Anyone in 4D from a Casual Monocular Video

    • It built a pipeline that reconstructs 4D humans, moving 3D people, by consistently generating videos from 16 viewpoints using only video from a single camera.
    • Reference-view compression (RCP) and viewpoint-group cycling (TCR) resolved the loss of consistency caused by increasing the number of viewpoints.
    • On the DNA-Rendering and DyMVHumans benchmarks, its 4D-reconstruction PSNR of 24.15/23.28 substantially exceeded the previous best method's 20.55/19.86.
    • It uses only a 3D skeleton as conditioning instead of unreliable depth information, allowing it to work on videos with unknown camera parameters.
    • It built the MVGameHuman training dataset using its own game engine, achieving generalization robust to varied outdoor videos.
    概要PDF
  23. 960AgentsarXiv:2602.11865

    Intelligent AI Delegation

    • The paper distinguishes task decomposition, which divides work into smaller pieces for allocation, from intelligent delegation. Intelligent delegation also transfers authority and responsibility, clarifies roles and boundaries, aligns intent, and establishes mutual trust.
    • The proposed framework has five pillars: dynamic assessment that continuously monitors the other party's current state, adaptive execution that revises plans as circumstances change, structural transparency that makes processes and outcomes inspectable, scalable market coordination that connects many participants at low cost, and system resilience that prevents cascading failures.
    • The central principle is contract-first decomposition, which means first breaking work into verifiable units. Subtasks whose outputs are difficult to verify or overly subjective should be decomposed further until they admit checks such as unit tests or formal proofs.
    • Selecting a delegate requires separating publicly visible reputation from trust criteria defined by the delegator. Authority should be limited to what the task requires and narrowed further under subdelegation, with progress monitoring, completion verification, and an accountability trail recording who delegated what to whom.
    • The paper argues that MCP lacks mechanisms for fine-grained authority delegation, A2A has no field for attaching proof of completion, AP2 lacks conditional settlement, and UCP is designed for commerce and therefore needs extensions for general task delegation. The proposal is conceptual and is not implemented or experimentally validated in the paper.
    概要PDF
  24. 959AI / GeneralarXiv:2608.16578

    Physics of Agents: Statistical Mechanics Predicts Collective Behavior of AI Agents

    • The study simulated around 10,000 communities of agents across objective mathematics questions and subjective political statements, and released trajectory data at both the individual and group levels.
    • Collective dynamics collapse into three regimes: indifference, polarization, and consensus. Interaction builds conviction and moves communities toward more ordered states.
    • On objective questions, communication improves collective accuracy: initially incorrect majorities switch to the correct answer more often than initially correct majorities switch to an incorrect answer. On subjective questions, systematic ideological drift can emerge, including movement toward the political right.
    • An extension of the Ising model, with intrinsic-field and interaction terms, predicts individual trajectories from initial opinions alone, outperforms all standard baselines, and generalizes to unseen community graphs.
    • The fitted parameters explain the observations: communities operate below the critical temperature, consistent with conviction buildup; attractive ties are stronger than repulsive ties, favoring consensus; and agents holding the correct answer exert the strongest pull, driving truth-seeking.
    概要PDF
  25. 958RoboticsarXiv:2608.18948

    RoboEdit: Turning Human Manipulation Videos into Scalable Robot Experience

    • The key design choice is to edit human videos directly rather than generate them anew, preserving the original scene context and physical consistency.
    • Cross-embodiment adaptation modules allow a single editor to cover the appearance, kinematics, and contact dynamics of seven robot embodiments.
    • A 3D Robot-State Decoder recovers the robot hand state in every frame, providing structured supervision for downstream learning and control.
    • The automatic RoboEdit-ADC pipeline uses depth normalization and physics-based refinement to reduce artifacts such as object interpenetration, floating contacts, and temporal jitter.
    • The authors built RoboEdit-14M, a dataset of 174,000 paired videos comprising 14 million frames across seven robot embodiments, and showed that it supports training real-world manipulation policies.
    概要PDF
  26. 957AI / GeneralarXiv:2608.19072

    What is Missing from AI Post-Training AI: An Empirical Analysis

    • The paper separates execution-level capability, meaning local refinement within a chosen strategy, from strategy-level capability, meaning revision of high-level decisions as evidence accumulates, and identifies the latter as the bottleneck.
    • The training strategy becomes fixed at the outset, and the entire remaining budget is spent on local adjustments. An agent can spend more than 10 hours iterating efficiently within the wrong strategy.
    • Strategy fixation reflects the agent's prior knowledge rather than the task itself. The same agent chooses similar strategies across different tasks, while different agents choose different strategies for the same task.
    • An experience scaffold combining an experiment log, a skill library, and an evaluator agent improves execution by 12.6 points on GSM8K and 40.8 points on HumanEval, but does not change the selected strategy.
    • Human corrections can redirect the strategy only during a brief window before training begins; once training starts, the agent returns to local adjustment loops. Increasing inference compute by two to eight times yields almost no benefit on the hardest task.
    概要PDF
  27. 956ResearcharXiv:2608.19068

    A metric on $S^2 \times S^2$ with positive sectional curvature

    • It proved by construction that the product of two spheres, S²×S², admits a metric with positive sectional curvature everywhere.
    • Starting from the standard metric, it applied a Cheeger deformation to obtain a nonnegatively curved Cheeger-Müter metric, then added a third-order perturbation.
    • Using the structure of the zero-curvature planes remaining at each generic point and the vanishing first variation of the perturbation, it made even the last flat directions positive.
    • It resolved a longstanding existence problem in differential geometry, expanding the concrete examples of positively curved 4-dimensional spaces.
    • Several calculations in the proof rely on MATHEMATICA, and accompanying code is provided, making independent verification of the symbolic calculations and execution of the code important for confirming the result.
    概要PDF
  28. 955Software EngineeringarXiv:2608.18933

    SkillForge: Self-Distilling Agents for Project-Specific Issue Resolution

    • SkillForge takes a proactive approach: it synthesizes issues directly from the repository and accumulates knowledge in advance, without relying on issue history or incurring costly per-issue test-time exploration.
    • The synthetic issues are test-covered bugs that actually execute, so the learning signal is verified through program behavior.
    • The distilled knowledge is organized in a two-tier store of global diagnostic skills and local intervention skills, then used through global initialization and just-in-time injection during issue resolution.
    • On SWE-bench Verified, SkillForge reaches 72.2% Pass@1 with DeepSeek-V3.2, an improvement of 5.8 percentage points, and 60.6% with GPT-5-mini, an improvement of 5.6 percentage points. On SWE-bench Pro, it improves the two models by 5.8 and 4.1 percentage points, respectively.
    • The type distribution of the synthetic issues does not overlap with that of the real evaluation issues. This indicates that the method targets new model-induced vulnerabilities rather than recycling the evaluation set.
    概要PDF
  29. 954Machine LearningarXiv:2608.18374

    Inference and Uncertainty Quantification for Streaming $r$-PCA

    • It removed a persistent remainder term from Oja's algorithm's error bound, obtaining a sharp operator-norm rate for general rank and sub-Gaussian data.
    • It proved lower bounds matching up to logarithmic factors in both dense-tail and sparse-tail covariance regimes, supporting the optimality of the convergence rate.
    • By linearizing the iterates, it derived a high-dimensional Gaussian approximation for the general-rank error and an explicit limiting covariance.
    • It proved the consistency of an online multiplier bootstrap, enabling estimates and uncertainty intervals to be computed together in a single pass through the data.
    • The current theory is limited to a constant learning rate, leaving extensions to varying learning rates and data with sparsity, Markov dependence, or quantization open.
    概要PDF
  30. 953LLMs / NLParXiv:2608.19197

    SPADE: Self-Play in Adaptive Synthetic Executable Environments

    • SPADE makes environment design itself a learnable component. The distribution of environments produced by the designer coevolves with the agent's capabilities.
    • A hint-based regret signal is the key mechanism. A purely adversarial designer proposes unsolvable tasks, while a purely cooperative designer proposes tasks with nothing to teach. The performance gap with and without hints automatically targets environments that are both feasible and worth learning from.
    • Because each environment is written as executable code with a Gym-style interface, one interface can cover both single-turn reasoning problems and multi-turn tool-use tasks.
    • Grounding the designer in documents from a pretraining corpus and maintaining an accumulated memory of previously created environments were both critical to performance.
    • At the 30B scale, SPADE improves over the strongest fixed-environment baseline by an average of 5.3 points across eight mathematics, science, coding, and reasoning benchmarks, by 5.7 points on BFCL v4 multi-turn, and by 13.9 points on ACEBench-Agent.
    概要PDF
  31. 952Machine LearningarXiv:2608.19098

    Open-MOPD: Diagnosing and Fixing Capability Imbalance in Multi-Teacher On-Policy Distillation

    • Even oracle routing based on ground-truth domain labels, which assigns each prompt to the correct teacher, fails to integrate the capabilities successfully. The bottleneck is capability integration itself, not routing.
    • Teacher disagreement was not the main cause. The problem was a distorted token budget: instruction-following tasks made up 20.3% of input prompts but received only 0.99% of gradient tokens.
    • The study separates the distortion into three causes: differences in response length across domains, different convergence rates across domains, and stale student reward signals caused by rollout reuse.
    • Combining token-share balancing, gap-aware dynamic budget allocation, and student reward refresh raised the improvement recovery rate from 35.6% to 83.4%.
    • The authors released the full training recipe, training trajectories, and evaluation suite, all reproducible on an academic setup with eight A100 80GB GPUs.
    概要PDF
  32. 951Computer VisionarXiv:2608.18388

    Depth Anything V4: Dynamic 4D Scene Reconstruction via Riemannian Flow Matching on 4D Gaussian Splatting

    • [Withdrawn paper] The latest arXiv version, v2, states: “This paper has been withdrawn by Jiaming Fan”.
    • The official version history dates the withdrawal to 20 August 2026, and the stated reason is “Major errors in research”.
    • Previously described methods and performance figures were pre-withdrawal claims and are not presented as valid research findings or grounds for recommending adoption.
    • Official withdrawal notice and version history: https://arxiv.org/abs/2608.18388 .
    概要PDF
  33. 950ResearcharXiv:2608.19095

    Quantum Magic in High Energy Collision

    • To address the dependence of quantum magic on the basis used to measure spin, it computed both QED and electroweak processes in the lab and helicity bases.
    • It found opposite trends: the helicity basis generally produces less magic in the ultrarelativistic regime, whereas the lab basis generally produces less in the nonrelativistic regime.
    • In a quantum-circuit implementation of ultrarelativistic Bhabha scattering, lab-basis circuits at general scattering angles required more T gates, demonstrating the connection between magic and computational cost.
    • In both bases, nature's weak mixing angle lay close to the value minimizing magic, showing that the earlier conclusion was not merely an artifact of basis choice.
    • The analysis is limited to event-by-event fixed bases and selected tree-level lepton-scattering processes, leaving extensions to higher-order corrections, non-Abelian interactions, and the full Standard Model for future work.
    概要PDF
  34. 949RoboticsarXiv:2608.17512

    Embodied-Navigator: Point, Think, Memorize, and Align for Efficient Navigation

    • The Pixel-to-3D Action Formulation lets the VLM operate through familiar 2D visual prompts, avoiding the spatial hallucinations and poor sample efficiency associated with learning 3D geometric transformations.
    • Selective reasoning activates only at critical nodes and compresses the remaining trajectories into space-time indicators, reducing both reasoning latency and memory overflow.
    • Two-level GRPO combines global outcome rewards with fine-grained process rewards to provide dense supervision and teach the model when to trigger reasoning based on the current state.
    • On R2R-CE, it achieved state-of-the-art performance with a 66.2% success rate while requiring only 90,000 training trajectories.
    • When deployed on a physical robot without additional tuning, it achieved a 60.0% success rate, outperforming StreamVLN at 49.0% and DualVLN at 53.0%.
    概要PDF
  35. 948AI / GeneralarXiv:2608.17528

    Agent Lightning v1.0: Towards Harnessed Agentic RL

    • In harnessed agentic reinforcement learning, the deployment harness, rather than the training engine, owns the environment-interaction loop, while the trainer observes only LLM request-response pairs. OpenClaw, Claude Code, and OpenHands are cited as examples of such harnesses.
    • The paper identifies four main challenges: retokenization, where identical text cannot be merged when token boundaries differ; a variable number of samples per rollout; loss normalization; and backend scheduling.
    • Computing both advantages and normalization at the rollout level rather than the sample level performed best, reaching a verification reward of 38.2%.
    • After the SWE-smith curation pipeline retained just 6,000 examples, reinforcement learning alone improved SWE-bench Verified performance from 41.8% to 56.4%.
    • Safeguards against reward hacking, including blocking Git commands and allowlisting network access, were necessary to stabilize training.
    概要PDF
  36. 947Computer VisionarXiv:2608.17832

    GenRec: Knowing Where to Reconstruct and Where to Generate

    • Training observed and unobserved pixels under a single loss makes reconstruction, which has one correct target, interfere with generation, which admits multiple plausible outputs.
    • A monocular depth estimator warps source pixels into the target view to construct an observation mask, which then controls the architecture, loss, and gradient flow.
    • A flow-matching backbone jointly reconstructs RGB and scene-coordinate maps, preserving cross-view consistency in a single sampling pass.
    • On two-view interpolation in RealEstate10K, it reached a PSNR of 19.80, well above the strongest baseline at 15.70, and achieved the best result on every metric in Mip-NeRF 360.
    • Inference takes about 11 seconds, more than two orders of magnitude faster than the strongest baseline at roughly 1,200 seconds.
    概要PDF
  37. 946Computer VisionarXiv:2608.17975

    aDSL: Agentic 3D Creation via Joint Agent-Program Design

    • The work starts from the observation that LLMs often miss precise numerical placements, such as a leg's exact coordinates, but reason well about structure, such as placing four chair legs beneath the corners of a seat.
    • aDSL emphasizes compositionality and spatial-relation operators, so agents do not have to manipulate low-level numerical values directly.
    • Disabling the self-correction loop reduced the visual question-answering score to 61.53 and also lowered the execution success rate, showing that iterative verification is essential.
    • Removing the planning stage increased the average number of revision rounds from 4.25 to 5.58, showing that structural decomposition determines efficiency.
    • An interactive front end, including an OpenClaw demonstration, supports iterative editing, while integration with an external high-fidelity generator preserves structure and improves detail.
    概要PDF
  38. 945Machine LearningarXiv:2608.17286

    Abra: Scaling Diffusion Image Training

    • A controlled model family spanning 10^19 to 10^22 FLOPs was used to measure the compute-optimal point for text-to-image diffusion models.
    • The optimum is about 200 image tokens per parameter, ten times the Chinchilla rule for language models at 20 tokens per parameter.
    • Diffusion models are robust to overtraining, so when compute remains available, using more data is preferable to increasing model size.
    • Generative-quality metrics such as FID and CLIPScore, representation quality, and optimal CFG settings also vary predictably with compute.
    • The compute-optimal token count rises with resolution, making high-resolution training compute-bound rather than data-bound.
    概要PDF
  39. 944Machine LearningarXiv:2608.17981

    Recirculation

    • Transformers pass representations forward through their layers, so shallow layers cannot revise their interpretations using context-sensitive representations formed in deeper layers. Recirculation feeds those deeper representations back down, helping the model track state.
    • Without changing the model weights, and with only modest hyperparameter tuning of the adaptive transform, the method reduced perplexity by 23% and increased GSM8K accuracy by 21% across the Gemma 3 family.
    • The GSM8K error rate fell by 8.8% at pass@1 and by 20.9% at pass@128.
    • During generation, the two stacks can run in parallel with almost no additional cost. Prefilling, however, must process each token sequentially and may become expensive for long contexts.
    • The authors present this as an approach that draws clues for structural improvement from the properties of a trained model, rather than imposing an entirely new architecture.
    概要PDF
  40. 943AI / GeneralarXiv:2608.17271

    ASI-Bench: At the Dawn of Artificial Superintelligence

    • The benchmark evaluates the same task under four levels of guidance: providing the full procedure, naming only the method, allowing autonomous method selection, and adding distracting information. This separates instruction following from autonomous research ability.
    • Across 18 agent and model combinations, the average score fell from 50.91 with the full procedure to 29.10 when only the method was specified and 26.62 when method selection was autonomous.
    • Scores varied substantially by harness, even for the same model. MiMo V2.5 Pro rose from 16.17 with its own harness to 23.25 with the Claude Code harness.
    • Providing only the method name was the most expensive condition, consuming an average of 6.91 million tokens per task. Incomplete guidance increased the cost of exploration.
    • The best-performing combination scored 51.60%, while a lower-cost combination delivered similar performance at one quarter of the cost, revealing a wide spread in cost efficiency.
    概要PDF
  41. 942RoboticsarXiv:2608.18077

    Hydra-0: Action Flow for Generalist World Modeling and Control

    • When robot commands are represented as point trajectories in the image plane and supplied as conditions, the world model can predict the effects of movement without knowing each embodiment's coordinate system.
    • The researchers trained one general-purpose world model by expressing heterogeneous videos in a common format, including human first-person demonstrations, hand-gripper footage, and dual-arm and single-arm robot videos.
    • In forward prediction, the method reduced robot-motion error by 90.4% and object-motion error by 60.2%. Its reproduced success rates for five RoboLab policies had a correlation of 0.96.
    • In reverse mode, the model observes only object motion from a human demonstration and generates a compatible robot movement. A trained action head then converts that movement into executable commands.
    • The study evaluated policies only in an open-loop setting and identified closed-loop evaluation as future work.
    概要PDF
  42. 941Machine LearningarXiv:2608.17310

    Agentic ESOpt: Fine-Tuning Long-Horizon LLM Agents with Minimal GPU Requirements

    • Using evolution strategies that evaluate random parameter perturbations through forward passes, it adjusted all parameters with only inference-level GPU memory.
    • It avoided the reinforcement-learning credit-assignment problem, which becomes harder as interactions lengthen, by attributing a long trajectory's reward to a single policy perturbation rather than distributing it across tokens.
    • On 15-turn Sudoku, it exceeded a strong GRPO baseline by 12.5%, and on ReAct mathematics and document question answering it improved over GRPO by an average of 8.3%.
    • It fully fine-tuned a Qwen3.5-27B web agent on four H100s, and a heuristic search that changed parameters and prompts together surpassed the baseline in 28 of 36 settings.
    • It requires additional tuning values such as perturbation scale, and the many forward passes become a disadvantage when environment evaluation is expensive; the issue of drifting in irrelevant directions during continual training also remains unclear.
    概要PDF
  43. 940ResearcharXiv:2608.16753

    Mathematics in the age of AI

    • Taking the premise that AI can perform research-level mathematics without debating AI capability itself, the paper argues that the mathematics community should explicitly state the goals and values it has long advanced implicitly.
    • In the second batch of the First Proof project, seven of ten problems received an effectively perfect passing assessment from at least one AI system.
    • The compute cost of these evaluations ranged from tens to hundreds of dollars per problem.
    • The paper warns that AI is adept at finding gaps between measurements and their intended targets, making the risk of Goodhart's law especially acute: once a measure becomes a target, it ceases to be a good measure.
    • In an age of abundance in which proofs are plentiful, the authors propose shifting cultural emphasis from generating proofs to digesting them through explanation, review, and incorporation into the canon.
    概要PDF
  44. 939LLMs / NLParXiv:2608.16647

    Every Coin Has Two Sides: On the Dual Nature of Generalization in On-Policy Distillation of Large Language Models

    • On-policy distillation reduces the difference between the teacher's and student's next-token distributions at each point in an answer generated by the student. The student therefore receives guidance in the contexts it actually visits during training.
    • For BigMath, the teacher selected 25,000 problems from each of three groups: problems answered correctly on all four attempts, problems answered incorrectly on all four attempts, and randomly selected problems. Across several teacher and student combinations, the three groups produced nearly identical final mathematics accuracy.
    • A teacher derived from the same base model brought the student close to the teacher's level in Chinese, long-horizon mathematics, coding, and science, even when training covered only English and short-horizon mathematics. A stronger teacher with a different base model produced much weaker transfer across those areas.
    • Changing the ratio of data from two teachers caused the student's scores across several areas to move together toward the original capability profile of the teacher used more heavily. For the first teacher pair, the average score on BeyondAIME and OlymMATH ranged from 25.1% to 27.1% across mixture conditions.
    • The experiments were limited to mathematics, coding, science, and instruction-following models. The multi-teacher study also covered only two teachers with fixed domain-based routing. The authors warn that domain partitioning should not be treated as a safety boundary because undesirable teacher behavior may also spread across domains.
    概要PDF
  45. 938Computer VisionarXiv:2608.16765

    TRACE-Bench: Decomposing and Diagnosing Multi-Reference Image Generation

    • The benchmark represents requests through four operations: object preservation, attribute disentanglement, attribute application, and scene composition. It uses the same formula structure to generate problems, score individual items, and trace the causes of failure.
    • The dataset contains approximately 1,600 evaluation cases, 631 formula templates, and about 4,000 reference images. Complexity ranges from one to eight operation slots.
    • The evaluation compared four commercial models with five open models. Nano Banana 2 achieved the highest average across the four operations at 0.8205, but scored 0.7384 on attribute disentanglement and 0.7989 on attribute application.
    • When 200 Emu3.5 cases were decomposed into simpler subproblems, 38.5% of all diagnostic outcomes were classified as interference arising when multiple references were composed together. Automated analysis and human failure localization agreed in 82.6% of cases.
    • Gemini-2.5-Pro scores the checklist used for the overall evaluation. When humans rechecked 400 outputs produced by two models across 200 cases, item-level agreement for the four scoring models ranged from 85.4% to 88.4%.
    概要PDF
  46. 937RoboticsarXiv:2608.16837

    HAF: Adapting Generalist VLAs to Humanoid Whole-Body Loco-manipulation via Hierarchical Action Flow and Spectral Latent RL

    • HAF-VLA activates head and locomotion, the waist, and bimanual manipulation sequentially across three stages. Stage indicators and the previous stage's KV cache connect earlier body movements with subsequent manipulation actions.
    • HAF-Steer compresses the VLA's temporal noise with a discrete cosine transform and learns only the first eight coefficients. It begins with behavior cloning, then applies SAC reinforcement learning that mixes offline data with real interactions while keeping the VLA backbone frozen.
    • The researchers tested seven tasks on TienKung 2.0 and 3.0, including loading laundry, carrying a basket, and throwing a ball. They collected 120 teleoperated demonstrations for each task and ran each method on each task ten times.
    • Across the seven tasks, HAF-VLA achieved an average progress score of 70.5%, compared with 53.3% for the strongest baseline, π0.5. It recorded the highest or joint-highest score on every task.
    • HAF-Steer improved HAF-VLA's success rate across all four conditions covering the original and new locations for two tasks. For example, toy storage at the new location improved from eight successes in ten trials to ten in ten. Its limitations include additional computational latency and dependence on the action range of the base VLA.
    概要PDF
  47. 936Machine LearningarXiv:2608.16760

    On the Principles Behind Neural Network Optimizers

    • Adam is neither always convergent nor always divergent, and the paper gives a safe region that depends on the problem and the number of mini-batches, with smaller batches requiring a larger second-moment coefficient beta2.
    • The Hessian of transformers becomes almost block diagonal during training, and the curvature differs greatly across blocks. The authors explain this as the reason Adam, with coordinate-wise learning rates, can be better than SGD.
    • They analyze simple neural networks and random matrix theory to show that consecutive large matrix multiplications create this Hessian structure, and this view also leads to a design that splits Muon correction across attention heads.
    • Adam-mini replaces more than 99.9% of the second-moment values with block-wise scalars, which cuts optimizer-state memory by 50%. By comparison, Adam needs about 104 GB just for the two moment states in a 13B-parameter model.
    • When training Llama 2 architectures from 39M to 1B parameters on C4, Adam-mini matches AdamW validation loss. However, this is a shortened version with no proofs, and rigorous analysis of attention and Mixture-of-Experts structures is left for future work.
    概要PDF
  48. 935RoboticsarXiv:2608.15875

    GigaBrain-0.7: Scaling Embodied Foundation Models to Emergent Capabilities with a Three-System Architecture

    • System 2 uses PaliGemma2 3B to understand the scene and task, System 1 uses PaliGemma2 3B plus a 0.5B action model to produce control values, and System 3 uses a 5B world model to predict future video and progress.
    • The trajectory data consists of 20,535.65 hours of real-robot data, 8,251.83 hours of UMI human demonstrations, 2,862.36 hours of first-person human video, 1,453.92 hours of simulation, and 4,153.22 hours of world-model generation.
    • In the System 3 ablation, the gift-wrapping success rate on AgileX PiPER-X rises from 0 percent for the base model to 80 percent when future video and progress are used together, and the clothing-folding completion time on AgileX PiPER falls from 107 seconds to 75 seconds.
    • Under the Co-Train setting on 50 RoboTwin 2.0 tasks, where 50 clean demonstrations per task and one policy are jointly trained, the results are 66.8 percent for Easy, 67.9 percent for Hard, and 67.35 percent on average, while pi0.5 reaches 70.7 percent, 46.0 percent, and 58.35 percent.
    • The real-robot tasks are usually evaluated with only 10 to 20 runs per setting and no confidence intervals, and the final MiMo score of 0.5704 also uses evaluation data that was included in training, so it is not an independent test result.
    概要PDF
  49. 934Information RetrievalarXiv:2608.15949

    Ask to Be Sure: Informative Interactions for Confident Multi-Turn LLM Recommendation

    • It computes uncertainty from the frequency and rank of movies in the recommendation list, and it uses how much the current response and the next user response reduce that uncertainty as the reward.
    • INSPIRED has 801 training dialogs and 99 test dialogs, while ReDial has 8,631 training dialogs and 1,036 test dialogs after filtering, and the simulated conversations run for up to five turns.
    • Averaged over three runs, turn-entropy DPO reaches 27.94% hit rate on INSPIRED simulated dialogs, which is higher than CollabLLM's reward at 26.60%, and on ReDial conversation-level entropy DPO reaches 32.83% versus 30.03%.
    • On ReDial, the average number of turns needed to reach the correct movie is 2.74 versus 2.86 for CollabLLM DPO, but this is not measured in real human conversations.
    • The evaluation depends on automatic recommendation metrics and LLM simulated users, sensitivity to the setting of five recommendation lists sampled five times is not validated, and direct recommendation Hit@1 remains low at 3.32% on INSPIRED and 2.24% on ReDial.
    概要PDF
  50. 933ResearcharXiv:2608.16360

    Contrastive Learning with Variational Regularization for Multi-Session EEG-to-Speech Decoding

    • The SpREAD dataset is EEG from one participant who listened to 1,353 Japanese sentences spoken by 18 speakers. Each sentence was measured three times on different days, for 45 sessions over 9 days.
    • The train, development, and test sets contain 3,195, 432, and 432 examples, respectively. Repeated measurements of the same sentence were always placed in the same split.
    • On the 432 test examples, the character error rate of 0.948 is significantly lower than the baseline 0.968. A classifier trained on the training set and evaluated on the development set reaches only 2.08%, which is no better than chance at 2.2%.
    • Using contrastive learning alone makes the acoustic correlation coefficient worse, dropping from 0.282 to 0.262. Adding variational regularization recovers it to 0.274, but that is still not better than the baseline, and speaker information improves only marginally.
    • There is only one participant, and the 18 speakers appear in both training and evaluation. The system handles only Japanese speech, and current EEG-based speech reconstruction remains far from practical.
    概要PDF