The paper introduces SGHA, a fully automated system that discovers research problems by structuring scientific literature into evidence-linked objects and a typed evidence graph. SGHA operates entirely on a local 9B open‑weight language model, avoiding proprietary frontier‑model APIs, and outputs traceable research‑problem families with assumptions, objectives, success criteria, and ambiguities. Comparative experiments in five machine‑learning domains show that SGHA’s corpus‑first, evidence‑constrained approach yields inspectable research‑problem formulation without relying on external models.
By Sarvesh Gharat, Junpei Komiyama
The paper investigates whether large language model (LLM) agents can design AI methods that outperform or differ from human-designed approaches. By mapping both human- and agent-designed methods into task‑specific algorithmic design spaces, the authors evaluate performance and algorithmic differences across multiple modalities. Results show that while agents occasionally match or exceed human state‑of‑the‑art performance, 96.8% of their designs fall within human‑derived spaces, often recombining or exactly matching existing human algorithms.
By Yikang Yang, Zhengxin Yang, Luzhou Peng, Minghao Luo, Yanqi Kan, Wanling Gao, Jianfeng Zhan
The paper introduces FACA, a Feedback‑Aware Credit Assignment method that aligns each agent reaction with the preceding user‑to‑user segment, computes a locally normalized reaction advantage, and adds it to the terminal outcome advantage without requiring an extra critic or rollout. Experiments show that FACA improves performance across nine domains, especially in Telecom, and maintains the same ordering of gains in zero‑shot benchmarks such as Pare‑Bench and Co‑Gym.
By Yiwen Zhao, Zhihao Wen, Yuchen Mao, Mingxuan Jiang, Yihao Hu, Pan Wang, Xin Zhang, Wei Wu
Agentic ESOpt proposes using evolution strategies (ES) instead of reinforcement learning to fine‑tune large language‑model agents for long‑horizon tasks. ES offers model scalability, flexibility, and better long‑horizon credit assignment, enabling full‑parameter optimization with minimal GPU memory. The framework samples parameter perturbations, evaluates agents with rewards, and updates online, achieving notable performance gains on WebArena‑Lite and in test‑time prompt‑parameter co‑evolution.
By Zhi Zheng, Rongsheng Chen, Yunpeng Ba, Zhenkun Wang, Yee Whye Teh, Wee Sun Lee
CityReal is a modular framework that uses large language model agents to simulate human-aligned urban behavior. It models agents as intention-driven decision makers who pursue coherent mobility and activity plans, learning habits and preferences over time. By training textual adapters to align agent decisions with observed population statistics, CityReal improves realism at both micro and macro levels and can scale to tens of thousands of agents for analyzing crowd density, place popularity, mobility flows, and well‑being under various urban scenarios.
By Nicolas Bougie, Xiaotong Ye, Narimasa Watanabe
The paper re‑evaluates memory‑based self‑improving agents by adding multiple runs to measure variance and by randomizing task order. It finds that agent performance is noisy in complex, multi‑step environments and that improvement depends heavily on the sequence of tasks, revealing a hidden curriculum effect. The authors suggest that underspecification of tasks and environments contributes to this fragility and demonstrate that adding detailed rubrics and feedback can partially mitigate performance drops, though gaps remain.
By Qinyuan Ye, Yu Li, Yada Pruksachatkun, Jiaxin Zhang, Chien-Sheng Wu
StartupBench is a new benchmark that evaluates general-purpose agents on end-to-end workflows derived from AI startup products that have proven market adoption. It translates real-world product workflows into deliverable-oriented tasks and assesses them with detailed rubrics. The study finds that even the best models complete only about 30% of these tasks, highlighting challenges such as complex instruction following and domain expertise.
By Liya Zhu, Xin Ma, Tao Liu, Haodong Wang, Ge Zhang, Jingzhe Ding, Qingshui Gu, Yongjie Zhong, Jinxiang Meng, Yuan Gao, Yunqiu Zhou, Hao Zhu, Jifeng He, Yongzhi Liao, Xinyi Zhang, Chaoxin Li, Yi Zhu, Xi Lin, Duju Zeng, Xiang Gao, Wen Zhang, Yunyang Wang, Duo Wang, Huan Zhou, Zuo Wang, Jin Chen, Kaiyuan Zhang, Chuqian Yu, Tianhao Yu, Longxiang Liu, Jianbo Xue, Huimin Che, Jiahao Wang, Yujia Qin, Jiaheng Liu, Shen Yan, Xiaolong Chang, Wenhao Huang
The paper introduces RATTL (Risk-Adversarial Total-Reward Learning), a framework that adjusts an agent’s caution based on epistemic uncertainty by using a Bayesian posterior over dynamics and a Wasserstein ambiguity set whose radius depends on that posterior. As evidence accumulates, the radius shrinks, smoothly transitioning the agent’s behavior from worst-case robustness to risk-neutral reward maximization. The authors prove a Safety Sandwich theorem showing RATTL’s value lies between the uninformed robust value and the full-knowledge optimum, and demonstrate the method on a binary-hazard example where the criterion reduces to Conditional Value-at-Risk.
By Deep Kumar Ganguly, Jan Kretinsky
The paper proposes using encoder‑decoder networks based on LSTM cells to reduce communication in distributed model predictive control (DMPC). Agents transmit a compressed representation of their messages, which receivers reconstruct, thereby lowering the data exchanged each time step. Experiments with mobile robot formations show that the trained networks maintain performance even when full communication is unavailable, and they can adapt to different prediction horizons without retraining.
By Torben Schiz, Pedro H. J. Nardelli, Henrik Ebel
The paper proposes EvalXRL, a benchmark that evaluates explainable reinforcement learning (XRL) methods by measuring how well their explanations help a large language model coding agent diagnose and fix bugs in RL agents. Unlike current metrics that focus on faithfulness or human ratings, EvalXRL uses a closed‑loop, scientific‑method style interaction where the coding agent repeatedly invokes XRL methods, refines hypotheses, and repairs the agent, scoring each method by the resulting RL reward. This approach enables a head‑to‑head comparison of multiple XRL techniques in realistic debugging scenarios.
By Ram Rachum, Yotam Amitai, B\'alint Gyevn\'ar, Reuth Mirsky, Cameron Allen
The paper introduces MemJack, a memory‑augmented multi‑agent framework that automatically generates jailbreak attacks on Vision‑Language Models (VLMs) using benign natural images as visual anchors. MemJack discovers visual anchors, camouflages them semantically, evaluates responses, repairs via reflection, and replans dynamically, forming a closed‑loop attack pipeline. The authors also create MemJack‑Bench, a dataset of over 113,000 interactive multimodal jailbreak trajectories, and show that MemJack achieves a 71.48% attack success rate against Qwen3‑VL‑Plus, reaching 90% under extended budgets, outperforming other baselines on natural‑image evaluation.
By Jianhao Chen, Haoyang Chen, Hanjie Zhao, Haozhe Liang, Zheng Wang, Tieyun Qian
The paper presents a theoretical analysis of the Rolling‑Horizon Collision Resolution (RHCR) framework for Lifelong Multi‑Agent Path Finding (L‑MAPF), proving its near‑optimality in a discounted MDP setting. Building on this, the authors introduce Group Decentralized RHCR (GD‑RHCR), which partitions agents via a transitive communication scheme and plans each partition in parallel, achieving similar optimality guarantees while reducing per‑plan computational cost. Experiments across various maps demonstrate that GD‑RHCR scales to higher agent counts with high throughput and lower cost compared to vanilla RHCR.
By Alex DeWeese, Jiaoyang Li, Guannan Qu
The paper argues that cross‑view correspondence, commonly used in agent evaluation and trace‑based learning, functions as a measurement intervention. Removing or altering this correspondence can create artificial sensitivity or invariance, and multiple optimal correspondences can obscure mechanism labels and learning credit. The authors propose a validity theory with two‑sided validation, all‑optima identification, and uncertainty propagation, and demonstrate through experiments that unvalidated correspondences can misattribute credit and erase harmful responses.
By Zhen Zhang, Ahmad Hafez, Amr Alanwar
The paper proposes Instant Episode Repetition (IER), a method that immediately repeats action sequences from high‑reward episodes during training to boost sample efficiency. Unlike passive replay techniques, IER actively shapes data collection by re‑executing successful behaviors for a set number of subsequent episodes. Experiments on MuJoCo, the DeepMind Control Suite, and a robotic manipulation task show that IER improves learning performance over standard SAC, TD3, and self‑imitation baselines.
By Hoda Yamani, Yuning Xing, Koen van Rijnsoever, Bruce A. MacDonald, Henry Williams
The paper introduces a small language model (SLM)-based framework that detects and classifies GNSS spoofing attacks on autonomous vehicles by converting vehicle states from GNSS and other sensors into structured semantic narratives. The SLM achieves performance comparable to large language models, with an average accuracy of 96.99%, while offering lower inference latency and reduced GPU memory usage. Field tests in Clemson, South Carolina, confirm the framework’s real‑time detection capabilities on resource‑constrained vehicular platforms.
By Abyad Enan, Sagar Dasgupta, Mizanur Rahman, Mashrur Chowdhury
Co‑RL is a multi‑agent reinforcement learning framework that trains several decoupled models without shared parameters, using rewards generated by their peers. By increasing cohort diversity—through heterogeneous model families, varying sizes, and rephrased training samples—Co‑RL reduces self‑reinforcing feedback loops, preserves behavioral diversity, and prevents training collapse. Across both text‑only and multimodal benchmarks, Co‑RL outperforms base models and prior label‑free methods, achieving gains of 3.0‑8.6% on seven text benchmarks and 2.3‑7.2% on four multimodal benchmarks, while matching or surpassing supervised approaches without any ground‑truth labels.
By Yunhao Yang, Yuexin Bian, Yunjie Tian, Di Fu, Tianjin Huang, Yuanyuan Shi, Ziang Xiao, Nuno Vasconcelos, Yijiang Li
The paper introduces ICD-Deepresearch, a workflow that combines foundation models for electronic health records (EHR) and language models with medical search and ICD dictionaries to forecast future ICD codes for upcoming clinical encounters. It evaluates candidate code transitions by linking patient evidence, external clinical relations, and exact code semantics within a fixed top‑K budget, using SparseEHR for initial priors, GPT‑5 for complementary forecasts, and a final selection step that validates, deduplicates, and ranks candidates. The method achieves patient‑averaged precision/recall of 24.60/35.09% on MIMIC‑III and 25.14/48.32% on MIMIC‑IV, with physicians rating 51–68% of its retrieved documents as useful, outperforming standalone GPT‑5 web search and Medical Deep Research.
By Junda Wang, Meysam Ghaffari, Akshat Choube, Mohsen Sharifi Renani, Hong Yu, Carlos Morato
The paper introduces PACE (Policy‑Attested Contract Execution), a framework that sits between large‑language‑model (LLM) based autonomous AI agents and on‑chain DeFi operations. PACE defines typed transaction intents, a deterministic policy verifier, and signed Policy Decision Records (PDRs) that cryptographically bind an approved intent, policy, and simulation report to the exact on‑chain execution bytes, providing replay and expiration protection. In evaluations across 40 tasks and six baselines, PACE achieves zero unsafe executions and zero false positives, outperforming unguarded agents by a large margin.
By Rabimba Karanjai (Larry), Yang Lu (Larry), Richard Williamson (Larry), Hemanth Hm (Larry), Prakhar Mehrotra (Larry), Lei Xu (Larry), Weidong (Larry), Shi
AdaLens is an interactive system designed to monitor and steer long-running, autonomous data analysis workflows powered by large language models. It provides a storyline-based interface that unifies analytical plans, execution progress, intermediate findings, and data-column involvement, enabling analysts to observe evolving reasoning and evidence. The system also offers steering interactions that allow users to redirect low-value directions or deepen promising ones during execution.
By Yangtian Liu, Yan Miao, Shuhan Liu, Yunfan Zhou, Dae Hyun Kim, Di Weng, Yingcai Wu
The paper "Token Optimization and Context Window Management in Multi‑Agent AI Workflows" introduces a practitioner framework that reduces token usage and latency in multi‑agent AI systems. It outlines six patterns—context stratification, fetch‑once/process‑locally architecture, schema‑contracted prompts, token‑aware fallback chains, semantic caching, and inter‑agent communication compression—and reports a 60‑70% token reduction and a 61‑116 second cold‑load latency improvement in production. A controlled study on relevance‑contrast context shows that mixing high‑ and low‑relevance items in prompts can improve relevance accuracy by up to +0.084.
whyItMatters":"The work provides concrete, repeatable engineering patterns that bridge research and production, enabling faster, cheaper, and more reliable AI workflows."
By Dvir Shamay