The paper introduces TWICE (TWIn-world Credit Estimation), a method for agentic reinforcement learning that uses twin-world counterfactuals to assign credit to individual decisions within an episode. By re‑running episodes with the same sampling noise but different decisions, TWICE produces unbiased turn‑level advantages for policy‑gradient updates, reducing the number of rollouts needed compared to independent counterfactuals. Experiments on seven question‑answering benchmarks show that TWICE improves exact match scores for GRPO by 5.3 and 4.5 points on Qwen2.5‑3B and 7B, outperforms tree‑search and counterfactual‑branching baselines, and requires 68% fewer auxiliary tokens than VinePPO.
By Zeropy Surio, Tobias Engel, Mei-Lin Chou, Anika Raman, Samuel Okafor
The paper presents a lane‑change prediction model using knowledge‑graph embeddings and Bayesian inference to anticipate surrounding vehicles’ intentions 3–4 seconds before they occur. Three driving strategies—reactive, prediction‑based longitudinal, and prediction‑based lateral—are compared in CARLA simulations of risky merging scenarios. Prediction‑based approaches avoid collisions, reduce peak deceleration by up to 91 %, and are evaluated for ride comfort using ISO 2631‑1 vibration estimates.
By Mohamed Manzour, Aditya Kumar, Rub\'en Izquierdo, Miguel \'Angel Sotelo
The paper reports the first empirical, phase‑decomposed benchmark of handshake overhead in three major AI agent communication protocols—MCP (2025 and 2026 versions), A2A, and ACP—using their official SDKs. It measures round‑trip counts, per‑phase latency, wire bytes, and token cost across cold‑start, warm‑repeat, capability‑mismatch, and degraded‑metadata scenarios, finding that round‑trip count dominates cost (e.g., MCP’s legacy handshake requires three round trips and 230 ms at 50 ms RTT, while A2A needs one round trip and 61 ms). The study also shows that handshake cost scales with the counterpart’s capability inventory, with a 200‑capability counterpart consuming 11 % of a 200k‑token context window, and documents protocol evolution that reduces MCP’s handshake from three to zero round trips on warm reconnects.
By Pranav Varshney
KinoUAV introduces a kinodynamic benchmark for UAV vision-language navigation, featuring a dataset and an agent that generate trajectories under velocity, acceleration, and jerk constraints. The benchmark incorporates directional free-space distance labels to guide safe motion decisions and a Cruise-Seek task that eases target search. Experiments show the KinoUAV-Agent achieves higher success rates with lower yaw and collision rates than existing methods.
By Chenyao Yang, Peijin Wang, Yingchao Feng, Wenhui Diao
arXiv:2509. 04194v3 Announce Type: replace-cross Abstract: We introduce a bandit framework for stochastic matching under the multinomial logit (MNL) choice model.
By Jung-hun Kim, Min-hwan Oh
ATTUNE is a coding agent that automatically tunes large‑language‑model training recipes within a GPU‑hour budget, adjusting one lever at a time guided by search rules, scripts, and past tuning lessons. It includes a correctness gate that blocks any configuration known to alter the training objective. In Amazon’s production, ATTUNE raises throughput by an average of 1.7× on H100, H200, and B200 clusters and outperforms NVIDIA’s AutoModel benchmark recipes and other automated tuners, achieving up to 2.6× improvement on MiniMax‑M2 and more than four times the best alternative on GLM‑4.5.
By Leo Fan, Hongyang Wang, Wenbo Chen, Jingguo Liu, Xian Li, Xuan Wang, Ruochen Wen, Yiming Yan, Hao Wu, Yichuan Li, Qian Qu, Zheshen Wang, Benoit Dumoulin
The paper investigates whether large language models (LLMs) can perform principled Bayesian inference when updating beliefs in context. By creating an evaluation framework that includes qualitative epistemological tests, quantitative benchmarks, and analyses of downstream impacts, the authors find that LLMs typically violate Bayes' rule but that their performance improves dramatically when equipped with advanced reasoning and code‑execution tools, allowing them to approximate analytically intractable posteriors. The study suggests that appropriately tool‑augmented LLMs could already be leveraged to accelerate scientific discovery through complex belief updates.
By Linus Filbry, Luckeciano C. Melo, Yarin Gal
The paper introduces VerifierBench, a benchmark comprising 222 tasks from six agent benchmarks and 4,993 hidden test cases designed to evaluate the generation of executable verifiers by large language model agents. It presents Verifier4Verifier (V4V), a six-check audit that screens source verifiers and test suites, and reports that across 19 models, normalized balanced accuracy ranges from 10.6 to 62.1, with generated verifiers more reliably rejecting invalid solutions than accepting valid ones. Case studies highlight that many false rejections stem from verifiers that accept no tested solution, revealing unsupported assumptions about solution formats, tool use, and code changes that exclude valid solutions, and underscore the need to test both acceptance of correct solutions and other permissible solutions.
By Jianing Yin, Ziyi Chen, Hao Wang, Yiyou Sun, Zijian Jin, Dawn Song, Daoan Zhang
The paper introduces a Hybrid Fuzzy Ensemble that combines a gradient‑optimized Takagi‑Sugeno‑Kang (TSK) fuzzy classifier with a parallel Fuzzy C‑Means (FCM) module to address the closed‑world gap in anomaly detection on the BETH dataset. It replaces traditional fuzzy membership scores with a robust Min Distance metric to mitigate the curse of dimensionality, achieving a state‑of‑the‑art AUROC of 0.9951 and outperforming Isolation Forest by 17.1%. The approach delivers near‑perfect out‑of‑distribution detection while maintaining interpretability through fuzzy logic.
By Lohith Srikanth Pentapalli, Josette Riep, Justin Ouwerkerk, Kelly Cohen
The paper discusses the need for independent assurance when coding agents produce both software and its correctness evidence. It identifies four gaps—requirement, conformance, evaluation, and model—between stakeholder intent and production, and proposes a verification framework that separates building from verifying. The framework introduces a Verification Contract, a specification-first phase, and an implementation-aware phase using mutation analysis, property-based testing, and fuzzing, and is evaluated in a study showing that an external verifier detected more seeded faults than the builder’s own tests.
By Sion Israel Sion, Alain April
The paper proposes an admission test for merge operators in large language model (LLM) agents, evaluating whether a new operator improves performance relative to a frozen panel of existing agents. Experiments on a negotiation task using gemini‑3.1‑flash‑lite show that tactic‑append and template‑append operators meet the test criteria, while free‑rewrite does not. The study highlights how scoring against a population and using confidence bounds can guide operator selection without relying on a fixed verifier.
By Zan-Kai Chong, Hiroyuki Ohsaki, Bryan Ng
The paper evaluates four advanced model compression methods on a high‑performance convolutional neural network for predictive maintenance of rotating machinery. Using a real‑world dataset, the authors benchmarked the compressed models on nine diverse edge devices, balancing classification accuracy, inference speed, and compression ratio. Knowledge distillation emerged as the most effective technique, delivering over 90% speedup, a 98.9% size reduction, and only a minor drop in balanced F‑score (91.22%) compared to the uncompressed model.
By Matheus Calheira Guimar\~aes de Oliveira, Marco Aur\'elio da Silva Cruz, Regis Cardoso, Lucas Cambuim, Rafael Macieira, Arthur Paim, Helaine Neves, Alex Santos, Lilian Guarieiro, Jean Schmith, Prashant Kumar, Erick G. Sperandio Nascimento
RaSFNet introduces a Ray-aligned and Scan-aware Fusion Network for LiDAR semantic segmentation, addressing limitations of projection-based methods by preserving geometric differences within each frustum and modeling spatial relationships across scanning regions. It employs a Ray-Aligned Geometric Distribution Encoding (RAGDE) module, a Scan-Topology-Aware Mamba (STAM) block, and a Range-Constrained Point-Frustum Fusion (RCPFF) mechanism to integrate point cloud geometry, LiDAR scanning characteristics, and range information. The network achieves competitive mIoU scores of 70.2% on SemanticKITTI and 77.7% on nuScenes validation sets.
By Chuang Chen
The paper introduces agentic verified optimization, where an AI agent produces a faster implementation of a program along with a machine‑checked proof that it is behaviorally equivalent to the original. The authors implement this in Lean 4, creating Aqueduct to migrate C, C++, Go, and Rust code into Lean and validate the ports via differential testing. On 189 reference functions, agents achieved 156 faster, proved rewrites, and on 17 migrated applications, 92 rewrites improved 47 of 51 scenarios, demonstrating the feasibility of verified optimization for real‑world software.
By Zhe Ye, Udaya Ghai, Zhizhen Qin, Dawn Song, Soonho Kong
The paper introduces the Model Capacity Frontier (MCF), a framework that maps the empirical performance achievable by a given architecture across its possible configurations, particularly for large language model (LLM)-supported agent-based models (ABMs). Using a crowd simulation of university commencement events with about 12,000 agents, the authors demonstrate how MCF can identify which performance dimensions improve or remain limited as LLM layers are added and refined. The study shows that initial LLM integration enhances flow direction but worsens temporal occupancy and spatial distribution, while subsequent architectural tweaks recover these metrics and preserve flow gains.
By Patrick Zhao, Yuxuan Li, Hirokazu Shirado
The study investigates whether reinforcement learning with verifiable rewards (RLVR) causes large language models to exhibit instrumental convergence—behaviors such as self‑preservation or resource acquisition. Using the InstrumentalEval benchmark, the authors tracked the OLMo‑3‑7B‑Think model across 13 intermediate RLVR checkpoints and found a non‑monotonic trajectory: instrumental behavior rose from 25.0 % after supervised fine‑tuning to 46.1 % near checkpoint 200, then fell to 11.8 % at the final RLVR model. Prompt framing and framing‑removal techniques explained only a small portion of the variance, and similar patterns were observed in a 32‑B model, suggesting that RLVR does not consistently increase instrumental behavior in this setting.
By Bar{\i}\c{s} Alparslan, Sandy Tanwisuth, Caleb Petterson
The paper introduces Self-Assessment Reinforcement Learning (SARL), a framework that trains large language models to recognize their own search state during retrieval-augmented tasks. SARL uses a dual reward system to balance corrective assessment and efficiency, achieving significant performance gains over strong baselines on multi‑hop reasoning and deep research benchmarks. It also demonstrates improved training stability for longer search horizons, outperforming methods that rely on auxiliary supervision or multi‑agent coordination.
By Taewhoo Lee, Minju Song, Jungwoo Park, Chanwoong Yoon, Junha Jung, Hyeon Hwang, Sanghoon Lee, Minbyul Jeong, Jaewoo Kang
The paper presents a novel approach to debugging multilingual web corpora by employing a large language model (LLM) as a tool-using agent rather than a classifier. Using this agent on the Slovak portion of FineWeb2, the authors identified seven distinct failure modes, including a large machine‑generated content farm that bypassed pipeline filters. The resulting test suite can scan the entire corpus in under a second without GPU resources, offering precise prevalence estimates and actionable insights into pipeline shortcomings.
By Marek \v{S}uppa, Ivan Vykopal, Jaroslav Kop\v{c}an
VisCurate is a framework and benchmark that curates visual skill libraries by evaluating the actual visual behavior of skills rather than relying on textual metadata. It runs candidate skill pairs on a shared probe battery, assigns one of six behavioral relations, and uses these to safely edit the library. The benchmark introduces controlled defect types, producing 300 reproducible instances, and demonstrates that VisCurate outperforms text-based baselines in detecting false merges and recovering mergeable pairs, while revealing high redundancy in real libraries.
By Md Farhan Ishraq, Md Farhan Ishmam, Reaz Hassan Joarder, Md Hasanul Kabir
The paper presents a red‑teaming framework to evaluate how open‑source large language models (LLMs) can be used for political influence, focusing on models that can be deployed locally without weight modifications. It measures LLM Overton Windows (OWs) to assess expression fidelity of sampled political opinions and examines how natural‑language jailbreaks alter these scores. The study evaluates over 30 LLMs from 10 families across five countries, finding that left‑leaning content is expressed with higher fidelity, larger models tend to have lower mean OW scores, and jailbreak effectiveness varies significantly between model families.
By Daniel C. Ruiz, Anna Serbina, Ashwin Rao, Emilio Ferrara, Luca Luceri