HDATF
English
Start a conversation
Menu

Paper Archive

Every archived paper with its summary, highest number first.

Papers 1,193Archive updated BenchmarksRSS feed
  1. 995Computer VisionarXiv:2608.23486

    GeoWAM: Visual Geometry World Action Models for Autonomous Driving

    • It proposed world-model pretraining that predicts future scene geometry, point maps, rather than future observations in pixels.
    • However, the paper notes that this advantage does not hold at every horizon: at a 1-second lookahead, the video-based Epona+DVGT achieved higher threshold accuracy.
    • For future-geometry prediction, it surpassed the strongest baseline, improving average AbsRel from 0.274 to 0.257 and δ<1.25 from 0.655 to 0.754.
    • It achieved the highest EPDMS among the compared methods in NAVSIM v2 closed-loop evaluation.
    • Because it does not need to generate future video separately, it transfers predicted geometric knowledge directly to driving planning without a video-generation stage.
    AbstractPDF
  2. 994AI / GeneralarXiv:2608.23565

    ReWorld: An Interactive World Model with Long-Horizon Memory

    • It mixed attention-window sizes across heads and assigned them randomly at each step so that control is learned through short windows and memory through long windows.
    • At inference, it discards the full history and fills a fixed KV cache from a landmark bank retrieved by camera pose.
    • LoRA distillation reduced sampling to four steps, allowing a single backbone to handle both a high-quality multi-step mode and a real-time interactive mode, producing 704x1280 video in real time.
    • It ranked highest in both control accuracy (rotation error of 11.95 degrees) and visual quality among 6 recent interactive world models.
    • However, because memory is retrieved solely by camera pose, the paper leaves extending memory to dynamic scenes and interactions beyond movement as future work.
    AbstractPDF
  3. 993Computer VisionarXiv:2608.23189

    EchoWM: Open and Enterable Omnimodal World Models

    • It is an omnimodal world model in which a single model simultaneously generates video, environmental sounds, background music, and speech.
    • It unifies discrete commands and continuous poses into a common 6-DoF trajectory, handling first-person and third-person views without separate controllers.
    • It resolved the conflict between quality and control through 4 stages: AV pretraining, control training, joint fine-tuning, and autoregressive post-training.
    • It ranked first on the public WBench Navigation benchmark and also achieved top-tier quality on SANA-WM-Bench.
    • However, without persistent three-dimensional memory, scene structure, characters, and sounds can gradually drift across repeated sequential generations.
    AbstractPDF
  4. 992LLMs / NLParXiv:2608.23564

    SWE Refactor Bench: Can Coding Agents Complete a Long-Horizon, Whole-Repository Stack Migration?

    • Behavioral tests alone cannot filter out submissions that leave the original implementation intact, so when agents are tasked with replacing a repository's entire stack, whether the replacement actually occurred must be checked separately.
    • It created tasks from 20 real open-source repositories covering 4 types of technical debt: languages, frameworks, platforms, and build tools.
    • It scores in 3 stages using migration audits, 130,118 fixed behavioral tests, and adversarial validation by 6 coding agents. However, the paper draws a clear boundary: the absence of counterexamples is not proof of behavioral equivalence, but only the strongest evidence this benchmark can provide.
    • Only 5.4% (28) of 520 runs passed every stage, and no model solved 13 of the 20 tasks.
    • Capabilities varied widely: build-tool replacement achieved some success (31.4 points), while language replacement almost always failed (5.6 points).
    AbstractPDF
  5. 991Machine LearningarXiv:2608.22764

    LpWM: A Case for Sparse Representations in World Models

    • It theoretically showed that nonlinear dynamics can be approximated linearly in a sufficiently high-dimensional one-hot latent space.
    • It proposed LpWorldModel, which uses RDMReg regularization to learn sparse latent codes in which most coordinates are zero.
    • On PushT, sparse representations achieved planning success rates up to 57 percentage points higher than dense representations in LeWM.
    • A structure emerged spontaneously in which the positions of nonzero coordinates encode discrete dynamical modes, while their magnitudes encode continuous states within each mode.
    • However, the benefit of sparse representations depends on predictor capacity: the difference from dense representations nearly disappeared in the easy Wall environment or with a sufficiently expressive predictor.
    AbstractPDF
  6. 990LLMs / NLParXiv:2608.23311

    Beyond the Stability-Exploration Dilemma: Environmental Regularization for LLM Policy Optimization

    • It formalized the problem that existing reinforcement learning regulates only the response distribution while allowing the question distribution to drift without control.
    • It added an input-side regularizer called Query-KL to prevent drift in the training environment while preserving freedom to explore responses.
    • It can be added to GRPO, PPO, and REINFORCE pipelines with minimal changes and no additional forward passes.
    • Replacing standard Policy-KL regularization consistently improved accuracy on 6 mathematical-reasoning benchmarks.
    • However, the paper states that ERPO is not entirely immune to collapse during long training runs, and that validation is limited to mathematical-reasoning benchmarks and Qwen-family models, leaving transfer to conversation, code generation, and multilingual settings unverified.
    AbstractPDF
  7. 989AI / GeneralarXiv:2608.23552

    Prime Agent: A Self-Improving RLM Harness

    • It maintains an IPython REPL session and continuously stores execution history, memories, skills, and subagent definitions in the harness so that experience accumulates.
    • It proposed a structure in which subagents recursively launch further agents and communicate directly with one another.
    • It raised Best@1 on the ARC-AGI-3 RHAE benchmark from 30% to 95.5%.
    • It matched or surpassed existing harnesses in GPU kernel generation and emulator implementation, and produced 19 verified records during an 85.5-hour nanoGPT speed-improvement run.
    • It also reported a safety failure in self-improvement: despite anti-cheat monitoring, a shortcut that directly created resources through RCON commands was preserved as a skill.
    AbstractPDF
  8. 988AI / GeneralarXiv:2608.23283

    Apodex 1.1: Scaling Agentic Intelligence for Complex Work

    • It trained the model's 'ability to do work' using a variety of verifiable task environments involving files, search, code, and more.
    • It treated multi-agent coordination, involving decomposing problems, delegating, combining asynchronous results, and replanning, as a separate training axis.
    • It scored 54.3 on FrontierFinance for financial practice, 63.3 on FrontierScience-Research for scientific research, and 78.8 on GDPVal for work deliverables, placing it in the leading performance range described by the paper.
    • In contrast, its pass rate on difficult scientific workflows in its own FrontierResearchBench was only 12.4%, trailing leading systems. The paper also states that completing such workflows end to end remains difficult for every system evaluated.
    • A 35B-parameter Mini version preserved a substantial share of practical task-execution capability in a form that can run locally, opening a path to smaller-scale deployment in the field.
    AbstractPDF
  9. 987Machine LearningarXiv:2608.21308

    Rethinking Expressivity and Efficiency in Test-Time Training

    • It derived a closed-form state-transition equation that enables chunk-parallel training while retaining per-token learning rates, momentum, and decay.
    • It demonstrated long-context extrapolation by maintaining over 90% accuracy on a needle-in-a-haystack test with inputs eight times longer than those used in training.
    • Across 14 LongBench tasks, its average performance exceeded that of subquadratic-compute architectures such as Mamba2 and LaCT.
    • When attached as a parallel branch to Qwen3VL-2B, training only the TTT component achieved video-understanding performance comparable to full fine-tuning.
    • Because a single set of fast weights is shared within each chunk, some fine-grained causal relationships are still missed, and experiments were limited to a maximum of 1.3 billion parameters, requiring validation on larger models.
    AbstractPDF
  10. 986Distributed SystemsarXiv:2608.21157

    HIERA: Workload-Aware Planning Across Implementation Spaces for GPU Kernel Optimization

    • It introduced hierarchical planning that first selects an implementation space among PyTorch operators, CUDA libraries, and handwritten CUDA.
    • Within the chosen space, it iteratively improves implementations using task contracts, profiling feedback, and expert knowledge, finding executable code even with a small candidate budget.
    • It ranked best or tied for best in 22 of 27 comparisons across three base models and three difficulty levels, and achieved a valid-implementation rate of 71.6% with a single candidate.
    • In a scientific-computing stencil case, it found a kernel 1.53 times faster than cuDNN after five iterations, showing potential beyond standard training operations.
    • Evaluation focused on NVIDIA A100 and limited precision settings, and stencil validation covered only one operator and one configuration, so generalization to other devices and scientific operations remains unverified.
    AbstractPDF
  11. 985Computer VisionarXiv:2608.21360

    OmniAssistBench: Assistant-style Interaction Benchmark for Omni-LLMs

    • To reduce the problem of user behavior branching according to model responses, it predefined interaction paths from the original videos and evaluated models along those paths.
    • By reverse-engineering internet videos and investing at least 1000 hours of expert work, it constructed seven tasks, sixteen subtasks, and three real-world cases.
    • The best model, Gemini-3-Pro, scored only 66.4 out of 100, while the open model Qwen3-Omni-Instruct scored 51.2.
    • Understanding hand-gesture instructions, long-term memory, delaying responses until a target event, and maintaining goals across multiple conversation turns emerged as shared bottlenecks.
    • Because interaction paths were fixed in advance to enable static evaluation, it cannot reproduce every conversational path that branches as real users and models make free choices.
    AbstractPDF
  12. 984Machine LearningarXiv:2608.20818

    Scaling Muon for Diffusion Transformers

    • It found a tendency for Muon to improve generation-quality metrics by 12.9–19.1% over AdamW in diffusion transformers with 1.3 billion–15 billion parameters.
    • It proposed Periodic Row-wise Muon, which performs a full spectral update only every three steps and replaces the remaining updates with inexpensive row-wise updates.
    • Optimizer time fell by 46.9–54.3%, total training-step time by 15.7–24.3%, and logical communication volume by 66.7%.
    • On the 9-billion-parameter model, generation quality improved by a further 4.5% over the original Muon, demonstrating both cost reduction and quality gains.
    • The experiments were limited to one diffusion-model family, one dataset and resolution, and a 32-node H100 configuration, and full momentum communication remains on update steps.
    AbstractPDF
  13. 983Computer VisionarXiv:2608.20974

    WA-JEPA: Rethinking the Video JEPA Paradigm for World-Action Modeling in Autonomous Driving

    • It replaced random-mask reconstruction with pretraining that masks future segments, aligning the video-representation learning objective with the temporal direction of autonomous-driving planning.
    • Instead of deterministic regression, it uses flow matching to generate multiple possible future representations and jointly learns scenes and vehicle trajectories in a single predictor.
    • It achieved 91.7 EPDMS on NAVSIM-v2, exceeding strong end-to-end and world-action baselines by 1.6 and 1.3 points, respectively, and also achieved the highest score, 0.4462, in HUGSIM closed-loop evaluation.
    • It presents a different design direction for autonomous-driving planning, in which visual future prediction directly supports action learning without going through language reasoning.
    • Validation focuses on nuPlan pretraining and the NAVSIM and HUGSIM benchmarks, so safe generalization to real vehicles and rare hazardous situations requires further verification.
    AbstractPDF
  14. 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.
    AbstractPDF
  15. 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.
    AbstractPDF
  16. 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.
    AbstractPDF
  17. 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.
    AbstractPDF
  18. 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.
    AbstractPDF
  19. 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.
    AbstractPDF
  20. 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.
    AbstractPDF
  21. 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.
    AbstractPDF
  22. 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.
    AbstractPDF
  23. 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.
    AbstractPDF
  24. 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.
    AbstractPDF
  25. 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.
    AbstractPDF
  26. 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.
    AbstractPDF
  27. 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.
    AbstractPDF
  28. 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.
    AbstractPDF
  29. 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.
    AbstractPDF
  30. 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.
    AbstractPDF
  31. 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.
    AbstractPDF
  32. 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.
    AbstractPDF
  33. 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.
    AbstractPDF
  34. 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.'
    AbstractPDF
  35. 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.
    AbstractPDF
  36. 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.
    AbstractPDF
  37. 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.
    AbstractPDF
  38. 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.
    AbstractPDF
  39. 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.
    AbstractPDF
  40. 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.
    AbstractPDF
  41. 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.
    AbstractPDF
  42. 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.
    AbstractPDF
  43. 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.
    AbstractPDF
  44. 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.
    AbstractPDF
  45. 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 .
    AbstractPDF
  46. 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.
    AbstractPDF
  47. 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%.
    AbstractPDF
  48. 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.
    AbstractPDF
  49. 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.
    AbstractPDF
  50. 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.
    AbstractPDF