Agent
-
MOSAIC: An Omni-Modal Benchmark for Long-Term Memory and Cross-Modal Reasoning in Conversational Agents
2026-10-21 05:55:20Hong Ting Tsang, Boyi Zhang, Haoyu Huang, Jiaxin Bai, Yufei Li, Yisen Gao, Zhongwei Xie, Hongyu Luo, Jeff Z. Pan, Yangqiu SongAbstract
Omni-modal assistants must remember not only what users say, but also what they show and convey. Existing long-term memory benchmarks are mostly text-and-image or report a single end-to-end score, so they cannot locate whether evidence is lost when it is represented, retrieved, delivered, or read. We introduce MOSAIC, a benchmark of 923 questions over persona-grounded histories of 11-24 sessions spanning text, image, video, and audio. MOSAIC constructs targeted spatial, temporal, and prosodic signals rather than decorative media, with the strongest causal validation for spatial information and selector-level validation for annotated prosody, and pairs them with a matched diagnostic protocol that freezes the question, reader, prompt, and budget while intervening on representation, retrieval, and delivery. Matched decomposition shows that representation loss is modality-dependent, residual reading dominates even under gold evidence, and abstention remains weak on these controls: strong answerable scores coexist with failures to refuse. MOSAIC thus turns omni-modal memory evaluation from a leaderboard into a diagnosis of write, retrieve, deliver, and read failures.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29249
PDF
-
Preference Is Not Permission: Personalizing the Auto-Purchase Line in Agentic Commerce
2026-10-21 05:15:31Ryotaro Shimizu, Sai HtaungKham, Mo FanAbstract
Personalized shopping benchmarks ask which item fits the user; we ask how far an agent may commit on that item, and pose learning the individual's auto-purchase line as a task. The boundary is evidence-conditioned and three-valued: whether committing is appropriate depends on the sufficiency of product information (warrant), reversibility, price against the user's norm, and urgency, not only on fit; and the natural output is don't buy, confirm first, or buy automatically. Across sixteen instruction-tuned models, autonomous commitment is misaligned with a human authorization reference in which fifteen annotators, role-playing the delegating principal from a given profile, choose the furthest action they would allow without confirmation: models commit even when fit is low and when their own evidence flag reports the warrant missing. This is overreach. The line is individual: these raters disagree, yet the disagreement is structured and partly predictable from a person's own other labels. A shallow learner on a few dozen of a person's own labels predicts their held-out auto-purchase line at a median of 0.79 by 30-40 labels, outperforming every evaluated prompt$\times$model configuration, including a per-annotator oracle over 80 configurations; the learned boundary does not transfer across domains. We release the counterfactual benchmark, human authorization reference, and annotation protocol after acceptance.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29216
PDF
-
NavBind: Diagnosing Instruction-Action Binding at Route Forks in Navigation VLAs
2026-10-21 04:50:26Josh Qixuan Sun, Xiaoying Xing, Yurun Chen, Huaiyuan Weng, Wentao Zhu, Mark CrowleyAbstract
Vision-language-action (VLA) navigation models translate visual observations and instructions into actions. Navigation success assesses whether an agent reaches the goal. However, reaching the goal successfully does not establish whether the model genuinely follows the instructions or relies on visual cues or navigation history. We introduce NavBind to diagnose instruction-action binding: whether changing the instruction influences the probabilistic distribution of the actions, with the navigation context fixed. We construct 3,252 route-fork contexts by recombining route-aligned instruction segments from R2R-CE and RxR-CE. Each context pairs full instructions that share the description of the route already traveled but specify different branches. Both are evaluated under the same agent state and navigation history. We compare generated local action sequences, the likelihoods of fixed reference action sequences under both instructions, and changes in visual attention and attribution. Across the evaluated models, higher reported navigation success does not imply that the model more faithfully follows the instructions.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29203
PDF
-
Reasoning-Search-Augmented LLMs: An Operational Taxonomy, Benchmark Analysis and Controlled Evidence Synthesis
2026-10-21 04:19:06Biswas Poudel, Nilson Chapagain, Amit Kumar, Xianshun JiangAbstract
Large language models (LLMs) demonstrate impressive emergent reasoning capabilities, yet they remain constrained by fixed model parameters and finite context windows. Even when augmented with static retrieval, they can produce statements that are unsupported by, or even contradictory to, retrieved evidence. This leads to hallucinations, difficulty incorporating up-to-date information, and failures to correctly interpret or reconcile sources. Recent work addresses these limitations by tightly coupling step-by-step reasoning with interactive search, treating retrieval as an action within an iterative agentic loop rather than a one-shot preprocessing step. This study reviews frameworks that implement such reasoning-search loops by interleaving planning, querying, retrieval, integration, verification, and stopping. We analyze these systems through four architectural axes: retrieval coupling, controller design, knowledge source, and grounding strictness. We separately treat the learning signal as a cross-cutting dimension because prompting, supervision, preference optimization, and reinforcement learning can instantiate the same architecture while producing materially different search, stopping, and grounding behavior. We further distinguish benchmark difficulty from RSA-loop demand and synthesize controlled within-study ablations and quality-cost deltas, avoiding invalid pooling across heterogeneous model backbones, retrievers, corpora, and evaluation protocols. Our goal is to clarify the landscape of reasoning-search-augmented LLMs and outline a unified perspective on the design choices shaping future research.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29189
PDF
-
LocalAgent: Collaborative Agentic Verification for Fine-Grained Instance-Level Consistency
2026-10-21 04:04:32Jiachen Guo, Xinshan Zhu, Boyun Wang, Yanyan Liang, Jun Wan, Feng Lin, Ajian LiuAbstract
Multimodal large language models (MLLMs) have made rapid progress in general visual understanding, yet they still struggle with fine-grained instance-level consistency verification, where the goal is to determine whether two images depict the exact same physical object under varying viewpoints, illumination, and backgrounds. This task requires both global semantic perception and precise examination of subtle local evidence, since semantically similar objects may differ only in small suspicious regions. To address this challenge, we propose LocalAgent, a collaborative agentic framework that integrates the complementary strengths of a generalist MLLM and lightweight specialist verifiers. The MLLM first reasons globally to identify suspicious local regions, which are then delegated to a Local Evidence Verifier (LEV) toolkit for fine-grained appearance and geometric matching. LEV comprises Local Appearance Consistency (LAC), which estimates appearance similarity under illumination and color changes, and Local Geometric Correspondence (LGC), which estimates spatial correspondence under viewpoint changes. LEV returns structured scalar evidence that the MLLM integrates for the final decision. We further introduce an auxiliary-capped reinforcement fine-tuning (RFT) strategy, in which answer correctness dominates auxiliary tool-use and format signals, enabling the policy to learn when verification is necessary. Experiments show that LocalAgent improves mean accuracy by 4.9% over the strongest CoT-RFT baseline while invoking a verifier on only 12.13% of samples and preserving the general multimodal capabilities of the underlying MLLM. The open-source repository is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29174
PDF
-
TrainBridge: Agentic NPU Training Optimization Guided by GPU Compilation Evidence
2026-10-21 04:00:48Yupeng Chen, Kunxiang Zhao, Xingjian Wang, Yue Zhao, Yijing Wu, Ye Liu, Kai Li, Congliang Chen, Ruoyu SunAbstract
We study how to improve training efficiency on Ascend NPUs. In the NPU training programs we study, framework-level compilation optimizations such as operator fusion are only partially applied, leaving fragmented computation and avoidable execution overheads. Profiling-guided tuning can address these inefficiencies, but it often requires costly trial and error. To make training on NPUs more efficient, we propose TrainBridge, an agentic approach for transferring GPU optimization guidance to NPU training. TrainBridge uses the highly optimized execution produced by mature GPU compilation stacks as a task-specific reference for NPU tuning. It extracts optimization intents from the GPU training execution to guide NPU program rewrites and further refines the rewritten program through iterative profiling. Our GPU-to-NPU migration experiments show that TrainBridge achieves an average 1.7x speedup across four workloads covering LLM pretraining, image classification, and vision-language training, and an average speedup of nearly 2.0x across 27 Modded-NanoGPT speedrun records with highly optimized GPU implementations.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29171
PDF
-
SimPT: An Automated Simulation Framework for Intervention-Based VLA Post-Training
2026-10-21 03:53:00Caicheng Wang, Zipeng Ye, Zhexuan Zhou, Zilong Huang, Xi Zhang, Yuchen XieAbstract
Human-in-the-loop (HIL) post-training improves vision-language-action (VLA) policies through expert corrections, but costly human supervision and repeated real-robot evaluation limit algorithm exploration and development. We introduce SimPT (Simulation-based Post-Training), a simulation framework that automates intervention-based VLA post-training by integrating policy rollouts, corrective data collection, policy training, and standalone evaluation. To automatically determine intervention timing and recovery strategies, we develop an Agent-Compiled Progress Graph that tracks task progress using privileged simulator state and retains stable snapshots during policy execution. After a rollout fails, a CuRobo-based expert attempts recovery from these snapshots, starting at more advanced task stages and falling back to earlier ones as needed to generate corrective trajectories. We implement and compare four post-training methods, RECAP, HG-DAgger, DPO+SFT, and FlowPRO, across seven RoboTwin manipulation tasks. Relative to a fixed-interval snapshot baseline, the Progress Graph reduces recovery attempts by 75.0% and increases recovery success from 93.4% to 98.9%. Using SimPT, we develop RECAP-CF-LPF, a refinement of RECAP that combines cross-fitting and low-pass filtering of value predictions. This refinement increases the overall success rate from 76.50% to 79.36% in held-out simulation trials. In real-robot HIL post-training, it increases the success rate from 72% to 96% across 25 scenarios of an Ethernet cable-insertion task. These results demonstrate the utility of SimPT for exploring and developing VLA post-training algorithms.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29161
PDF
-
VeeReview: Measuring Build-Verify Pairs in Agent-Authored Pull Requests from a V-Model Perspective
2026-10-21 03:25:44Gabriel Dautovi, Jie JW WuAbstract
Context: Reviewing a pull request requires verification evidence that addresses the proposed change. Coding agent benchmarks and studies of pull request outcomes do not directly assess this relationship. Objective: We test whether the V-Model's build and verify pairs can measure verification evidence in pull requests and support comparisons between agent-authored and human-authored changes. Method: VeeReview classifies four build and verify pairs using repository signals and a language-model judge. Three pairs use model judgements constrained by deterministic signals, and a rule over the diff and repository tree decides Module Design-Unit Test. We evaluated VeeReview against a 25-pull request reference set labelled by two raters and applied it to 1,454 matched pairs of agent and human pull requests from 741 repositories. Results: VeeReview assigned lower build-only rates to agent-authored pull requests on all four pairs, contrary to the preregistered hypothesis (mixed-effects odds ratios 0.52 to 0.86). Architecture-Integration was not significant. Standardization for linked issues, shipped tests and size reduced matched-rate differences by 34% to 60%, and the rule-decided pair's standardized interval included zero. The instrument did not meet its preregistered calibration criterion: agreement with adjudicated labels was 4% to 60% against a required 80%, and inter-rater agreement was weak (kappa = 0.154). Conclusions: The rule-decided pair shows that a process measure judged by a language model partly reflects how a pull request is presented rather than the verification work behind it. We report the cohort differences as observations made with this instrument, and we argue that such measures need a mechanically decided component and a validated human reference before they can support comparisons of verification practice.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29143
PDF
-
Agentic Video Understanding: A Survey
2026-10-21 02:49:27Xinyu Deng, Siwen Luo, Daochang LiuAbstract
As large language models (LLMs) become capable of processing increasingly diverse modalities and longer temporal contexts, an emerging line of work is moving beyond fixed video-language inference toward agentic systems that actively decide what information to inspect, retain, verify, and act upon. This survey reviews video understanding agents: systems that use video as the primary information source and solve understanding tasks through adaptive state construction and action selection. We first formalize an agent loop for video understanding, then address a central question: why do agents matter for video understanding? To answer this, we organize the literature through a challenge-to-design taxonomy, linking context bottlenecks to hierarchical evidence memory, evidence sparsity to active evidence acquisition, temporal causality to state and process tracking, and multimodal ambiguity to role-specialized coordination. We further review state space paradigms, learning paradigms, supervision signals, benchmarks, and evaluation protocols. Finally, we identify open directions toward agentic-native temporal modeling and video-native agents. Project page: this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2609.31713
PDF
-
Agentic Music Post-Production
2026-10-21 02:36:51Jaywon Koo, Junghyun Koo, Marco A. Martinez-Ramirez, Wei-Hsiang Liao, Seungheon Doh, Qiyu Wu, Vicente Ordonez, Yuki MitsufujiAbstract
We introduce MixAgent, an agentic pipeline for music post-production that follows the workflow of audio engineers to build interpretable and editable mixing plans guided by a reference song. A grouping agent first routes related input tracks, such as drum recordings, into subgroups. A planning agent then uses retrieved examples to choose and order audio effects, such as volume control and reverb, for tracks, subgroups, and the full mix. To refine the mix, a detector identifies what needs correction, and a critic proposes numerical parameter adjustments from tracks to the full mix. In our experiments, prompting recent multimodal models to listen and revise often worsens the mix, making reliable correction a central challenge. We therefore build MixCritic, a dataset of 45,657 repair critiques derived from parameters fitted to professional mixes, and train the critic with supervised fine-tuning followed by preference optimization on rendered audio outcomes. On MedleyDB, learned refinement reduces error by 19% relative to the initial mix, with preference optimization reducing it by 8.4% over supervised training alone. Retrieved examples further improve effects selection, and a human listening test shows that MixAgent gets higher quality ratings than mixes produced by Qwen3-Omni, Gemini 3.8 Flash and GPT-4o-audio.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29093
PDF
-
Our Own Jev: Typed Decisions in One Forward Pass, from an In-House Judge to Open LoRA Models
2026-10-21 01:28:58Yuchuan Weng, Chi Liu, Xin Chen, Srinivasan ManoharanAbstract
A typed decision maps evidence, a question and a fixed set of options to a probability for every option, computed in one forward pass, by reading a language model's logits or a small head over its hidden states, instead of generating and parsing text. The commercial system Jev popularised this System~1 interface, and open reproductions have followed. We report two uses of the pattern. \emph{Part I} describes an in-house System~1 judge that triages about 148K agentic coding session records for a training corpus. It replaces a frontier LLM judge that cost on the order of \$1,000 per pass and whose stored labels matched a fresh two-judge consensus on only 26.9\% of the sessions where the two judges agreed; a self-hosted frozen backbone with a rank-16 LoRA adapter and a pointer head answers three typed questions in one forward pass, judges 1,000 sessions in about three minutes of one GPU, about \$0.06 at a reference rate and roughly 1\% of the frontier judge's cost, and keeps every session inside the organisation. \emph{Part II} asks whether a readout that adds no parameters can be competitive and whether training one helps. Applying the public Eikos recipe and corpus of about 25K decisions to three further backbones (1B, 2B and 35B), we read the frozen output layer of a LoRA-adapted decoder. On the 111 public hard items of JevBench our single-seed 35B, developed with those items in view, scores 79.3\%, not significantly different from the commercial reference (73.0\%, unadjusted $p=0.12$) and ahead of five open systems after Holm--Bonferroni correction; our 2B is ahead of all four kev checkpoints on the leaderboard. A single-seed ablation finds no detectable benefit from training the 26 letter rows of the output layer, and a retrain on a later corpus release moves the 2B's hard-tier accuracy by 11.7 points, which shows how little small evaluations resolve.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29052
PDF
-
Long-MDR: Long-Context Reinforcement Learning for Multimodal Deep-Research Agents
2026-10-21 01:27:11On Tai Tang, Huanyao Zhang, Jiepeng Zhou, Wentao ZhangAbstract
The next generation of multimodal research agents must reason over long-lived research histories rather than short model completions. During a single task, an agent may repeatedly search the web, inspect visual evidence, revisit earlier hypotheses, and accumulate tens of thousands of tokens of multimodal context. Despite this trend, online RL for multimodal research agents remains largely confined to shorter contexts and interaction horizons. We push online RL training to 128k context and 75+ tool-interaction turns. To our knowledge, this is the first online multimodal deep-research RL study trained at 128k context, and the first trained with a 75 tool-turn horizon. Scaling to this regime exposes several practical limitations of conventional RL training. Early in training, weak policies make poor use of large interaction budgets, causing expensive rollouts with little reward improvement. Later, policy entropy can collapse before performance has saturated, prematurely ending useful learning. We introduce Long-MDR, a three-component training recipe designed specifically for this setting: On-Policy Distillation Warmup, Progressive Horizon Expansion, and Entropy-Triggered Rescue. Together, these techniques improve both the learning efficiency and stability of long-horizon RL, enabling continued gains in a regime where direct training is slow and costly. Code will be available at this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2610.05195
PDF
-
Resonance and Depth: Composing Timbral Identity and Spatial Form in an Artist-in-the-Loop Co-Creative Multi-Agent Percussive Performance
Abstract
This research-creation artistic report presents Resonance and Depth, a 9-minute immersive electroacoustic/acousmatic performance that treats timbre and space as primary compositional materials. Built from live electronic percussion and a personal corpus recorded, curated, and composed by the author, the work investigates how timbral identity and musical form can emerge through real-time recombination, spatialisation, and performer-system interaction rather than conventional harmonic development. The performance is realised as an artist-in-the-loop co-creative multi-agent system integrating MACAT, MACataRT, a MIDI-based musical agent, a spatialisation layer, and a conductor agent. These components recombine and transform the performer's material through concatenative synthesis, audio mosaicing, symbolic mediation, physically inspired synthesis, and object-based spatialisation. The work approaches spatialisation as an active compositional process in which position, trajectory, depth, diffusion, width, and apparent size participate in musical phrasing and form. Through this practice-based framework, Resonance and Depth examines how performer identity can be retained, transformed, and redistributed across human, algorithmic, timbral, and spatial processes within an immersive co-creative performance ecology.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29026
PDF
-
HybridResearcher: Bridging the Structured-Unstructured Boundary for Deep Research Agents
2026-10-21 00:41:07Soyoung Yoon, Peiran Xu, Han Wang, Ruofan Wu, Xiaodong Yu, Feng Yan, Mohit Bansal, Debiao Li, Seung-Won Hwang, Yuxiong He, Zhewei YaoAbstract
While LLM agents are increasingly capable of gathering and synthesizing evidence from a single homogeneous source, such as structured tables or unstructured data, real-world analytical questions frequently span heterogeneous sources. For example, answering one enterprise question may require both a result computed over an internal database and evidence found on the web. Recent evaluations on benchmarks like HybridDeepResearch show that even frontier agents struggle when evidence must cross this boundary. While these benchmarks have exposed the limitation, methodologies to explicitly teach cross-source reasoning remain absent. To address this, we introduce HybridResearcher, a 4B agent trained to bridge evidence across the structured and unstructured boundary. To build it, we synthesize cross-source bridge trajectories by starting with validated single-source tasks-such as existing text-to-SQL or web search queries-and fabricating the missing source around a bridge entity drawn from the seed's own output. We retain a sample only when strict construction gates confirm that neither side alone suffices, ensuring the data requires true cross-source interdependence. Finally, we train on this curated dataset using supervised fine-tuning followed by reinforcement learning. Evaluated on the existing HybridDeepResearch benchmark, HybridResearcher improves over our base model, Qwen3.5-4B, by 30.0 points pass@8 and surpasses the far larger Qwen3.5-397B-A17B by 8.3 points, while also improving over its base model on established, held-out single-source benchmarks (BrowseComp and BrowseComp-Plus), demonstrating that this complex cross-source reasoning can be taught directly to small models without triggering catastrophic forgetting.
Abstract (translated)
URL
https://arxiv.org/abs/2610.24899
PDF
-
VeriEvolve: Evaluator-First Evolution for Tool-Using Agents
2026-10-21 00:13:29Yueshen Li, Kamer Ali Yuksel, Gokhan Tur, Dilek Hakkani-T\"ur, Jo\~ao Pedro MaiaAbstract
A central obstacle to self-improving agents is the absence of a reliable fitness signal for realistic stateful tasks. We introduce \textsc{VeriEvolve}, an evaluator-first evolution framework that constructs this missing evaluation layer before optimizing the agent. From task instruction, VeriEvolve synthesizes instruction-grounded challenge states and an executable evaluator that checks them. During each round of workflow evolution, the evaluator is frozen and serves as the fitness signal. Between rounds, VeriEvolve revises the evaluator by converting the end states of strong and weak runs into a pair of counterfactual test cases. Across all 30 MCPMark Filesystem tasks, VeriEvolve constructs 30 evaluators with 141 challenge states, achieving 141/141 agreement on the generated suites and 113/122 agreement on held-out states produced by executor agents. Evaluator-guided evolution increases externally verified success from 6/30 to 18/30 for GLM-5.2 and from 27/30 to 30/30 for Claude; refinement removes all three false positives that the weak executor exposes. Cross-environment studies on PostgreSQL, Playwright, Notion, and GitHub further expose complementary limits from observability, state representation, and search/execution capacity. These results show that executable evaluators can be synthesized and used to guide search when task success is observable and an external verifier is available for calibration.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29005
PDF
-
FlashNext-RL: A System Optimization Study for Long-Horizon Agentic RL of Qwen3.8-Flash-Next
2026-10-20 23:53:09Zhewei Yao, Yuxiong HeAbstract
Long-horizon agentic reinforcement learning (RL) requires both efficient rollout generation and training on long sequences. However, the memory requirements and model-specific operators of Qwen3.8-Flash-Next make this difficult with existing open-source implementations. To resolve these limitations, we introduce FlashNext-RL, a model-specific RL training optimization framework based on prime-rl, with vLLM for software engineering (SWE) rollout generation. Our main focus is BF16 training of the full 48-layer model on a single 8xB200 node, with the first two layers, the n-gram table, the sparse-attention indexer and the multi-token prediction head frozen. We optimize host offload, gradient accumulation and the kernels for mixture-of-experts, sparse attention and the gated residual stream. In the full-model development study, non-padding training throughput increases from approximately 395 tokens/s/GPU in the initial 32K-token configuration to 4,438 tokens/s/GPU in the optimized 131K-token configuration, with hardware matmul throughput (aka without considering the cost of attention components) increasing from approximately 20 to 213 TFLOPs/GPU. A controlled cumulative study on an 8-layer model further quantifies the contributions of individual optimizations. Together with three inference nodes, the system supports a live SWE RL run with a target batch of 1,024 trainable rollouts and a 131,072-token context limit. Over steps 5-15, an iteration takes 22.4 minutes on average, with bounded entropy and trainer-sampler mismatch KL. These results highlight the importance of tailoring RL training to the model's architecture, memory layout and kernel requirements, rather than relying only on general-purpose RL infrastructure.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28998
PDF
-
One Policy, Many Styles: Where to Aggregate Preferences for Steerable Humanoid Control?
2026-10-20 20:59:05Ziqian Huang, Yuzhuo Chen, Shimin Zhang, Junyang He, Xinyue Annie Yang, Animesh GargAbstract
Modern robotic agents are increasingly expected to support steerable control, where a single policy can adjust its behavioral style at runtime under changing user preferences. Preference-conditioned reinforcement learning provides this interface, but conditioning alone does not ensure that a requested behavior change is strong, does not leak into unrelated behaviors, and does not compromise task performance. Our evaluation combines Preference Reward Responsiveness (PRR) with coverage and utility metrics. We use these measures to compare reward-level, advantage-level, loss-level, and PCGrad-based aggregation across two humanoid-control frameworks. No aggregation level performs best across all environments and criteria. Experiments show that aggregating objectives early produces a wider range of behavioral responses but greater unintended changes in other objectives, while aggregating them later improves control precision at some cost to responsiveness.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28925
PDF
-
Learning Dynamic Games for Cooperative Iterative Tasks
Abstract
We consider cooperative and iterative tasks involving multiple agents whose interactions are modeled as a dynamic game with shared constraints. We propose a general learning framework applicable to nonlinear and nonconvex dynamic games that uses previously executed feasible trajectories to construct a sampled joint safe set and player-specific cost-to-go functions. At each time step, the agents jointly select a previously sampled terminal state and a Generalized Nash equilibrium used to reach it. A new solution is adopted only if it satisfies a cooperative acceptance rule agreed upon by the players. The framework accommodates individual non-degradation requirements and selection criteria such as weighted costs and Nash bargaining. We show that the proposed framework guarantees recursive feasibility and characterize conditions under which each player's cost is non-increasing across iterations.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28901
PDF
-
Building AI Research Teams for Autonomous Discovery of Linear Attention Architectures
2026-10-20 20:09:31Yujian Liu, Li An, Zelin He, Jacob Helwig, Jiabao Ji, Rohit Jain, Siyu Zhu, Shiyu ChangAbstract
Autonomous agents can increasingly propose ideas, modify code, and run experiments, creating an opportunity to accelerate scientific discovery using teams of collaborative AI researchers. To escape local minima in the discovery, we propose to encourage orthogonal exploration by constructing the team using agents with varying expertise. However, realizing this design requires more than diverse backgrounds. When experiments are costly, a team can test only a limited number of ideas, thereby creating the need to balance exploration and exploitation when selecting a subset of researchers to propose and execute new experiments. Furthermore, each result should shape what the team tries next. We therefore ask how AI researchers with different scientific backgrounds can build on one another's findings and sustain progress under a limited experimental budget. To address these challenges, we build a team of persistent AI researchers, each grounded in a different discipline, who keep private research histories, share experimental evidence, and receive research opportunities through adaptive allocation. We demonstrate the effectiveness of this design by deploying the team to search for linear attention architectures, where it outperforms GDN2 at 1.3B scale on language modeling, common-sense reasoning, and synthetic and real-world retrieval tasks. In controlled comparisons, removing any single organizational component from our design yields markedly weaker discoveries: without disciplinary backgrounds, the search plateaus early; without private histories, researchers make less progress despite identical shared evidence; and uniform selection falls short of adaptive allocation within the same budget. These results suggest that sustained autonomous discovery depends not only on stronger agents, but also on who explores, what they remember, and how research effort is allocated.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28887
PDF
-
SANE: Automated Scorer-Specific Metamorphic Testing for LLM Judges
Abstract
LLM judges increasingly mediate benchmark evaluation, model selection, data curation, and training. Their decisions are useful only when each judge responds reliably to changes affecting its scoring criterion. Existing methods for reliability testing perturb inputs with known expected effects on judge scores. Prominent LLM-judge testing harnesses combine human-authored perturbation recipes with synthetic test generation, leaving scorer-specific recipe ideation outside the automated pipeline. This limits their use to fixed perturbations and pre-assumed scorer types. We introduce SANE, an agent-based framework that encodes behavioral requirements for stochastic evaluators as reusable agent instructions. Given a scorer specification, implementation, and representative inputs, the pipeline first identifies scorer-specific capabilities, then automatically derives perturbation recipes and expected score relations for each. SANE also defines a scorer-independent validation framework for generated judge tests. It determines whether each recipe agrees with scorer intent and whether each instantiated case realizes the recipe and supports its expected relation. Because SANE generates a recipe inventory tailored to each scorer, the number of aligned recipes grows as scorers are added. Across five domains, SANE produces 3.7 times as many validated test families as the Judge Reliability Harness (JRH) at comparable end-to-end validity and detects scorer deficiencies in every domain, including professionalism and toxicity, where JRH detects none. On faithfulness, SANE achieves strong human alignment in judge ranking: its test-suite scores rank 11 judge models consistently with human-labeled FaithBench performance (Spearman rho = 0.852), while JRH does not (rho = 0.100).
Abstract (translated)
URL
https://arxiv.org/abs/2610.28886
PDF
