FrugalEvo introduces a cost‑aware evolutionary framework that pairs a high‑cost, powerful LLM with a cheaper LLM to explore and refine program solutions, respectively. It incorporates a cache‑efficient evolution process that maximizes prefix reuse across steps and defines a Budget‑Aware Area Under the Curve (BA‑AUC) metric to evaluate solution quality within a fixed cost budget. Across ten mathematical and systems optimization tasks, FrugalEvo matches or surpasses state‑of‑the‑art baselines in final solution quality and achieves higher BA‑AUC on nine tasks, while delivering new state‑of‑the‑art performance on circle packing for costs as low as $0.55 to $1.68.
By Hui Chen, Xuan Qi, James Xu Zhao, Zhaopeng Feng, Shilong Liu, Kuang Xu, Pang Wei Koh, Bryan Hooi
The paper introduces a counterfactual evidence audit to assess how large language model agents respond to different sets of documents. By exposing agents to two mirrored sets of five documents and measuring changes in six downstream decisions, the authors predict how the agents will behave when presented with a full 45‑document context. Across multiple open‑weight model families and tasks, the five‑document audit accurately forecasts full‑context effects, reducing prediction error by 62% and correctly identifying the direction of most material effects.
By Rana Muhammad Usman
The paper proposes a framework for agent calibration that ensures consistent capabilities when agents are deployed, migrated, or scaled across different models, harnesses, infrastructures, and user contexts. It defines standards for basic capabilities, technical environments, and user contexts, diagnoses gaps, applies revisions, and rechecks within fixed budgets, using diagnostic source behavior and mandatory tests to qualify agents. The framework includes semantic checkpoints, repair localization, and reusable configuration artifacts, and outlines an automatic calibration tool that relies on limited authorized user trajectories and tests, though empirical validation remains pending.
By Yaxiao Liu (PwC China AI Center), Pengbo Liu (PwC China AI Center), Yiwen Liu (PwC China AI Center), Yihua Guan (PwC China AI Center), Jiaxing Song (Tsinghua University)
The paper presents a systematic study of token consumption in AI agents performing coding tasks. It finds that agentic tasks are far more expensive—about 1000 times more tokens than code reasoning or chat—primarily due to input tokens, and that token usage varies wildly, with accuracy peaking at moderate costs. Models differ significantly in efficiency, and current frontier models cannot reliably predict their own token usage, often underestimating it.
By Longju Bai, Zhemin Huang, Xingyao Wang, Jiao Sun, Rada Mihalcea, Erik Brynjolfsson, Alex Pentland, Jiaxin Pei
MACTS-EM is a new multi‑agent framework for time series forecasting that combines domain‑specialised agents, a meta‑cognitive allocation layer, emergent memory for cross‑domain transfer, multimodal context integration, and adversarial robustness. The authors evaluate the system on financial, climate, energy, and pandemic data, reporting 8‑12% higher accuracy, 22‑27% better zero‑shot transfer, 16‑21% greater resilience to regime shifts, and 15‑18% faster recovery from distribution changes compared to existing methods. These results suggest that collaborative, agent‑based approaches can outperform traditional architectures in complex, real‑world forecasting tasks.
By Ahmad Shahi, Mamehgol Yousefi
SCAD (Structured Credit Assignment and Distillation) is a new method for training long‑horizon agents that separates planning from bounded subtask execution, distills execution locally, and refines planning credit using cross‑rollout subtask prefix trees. It gives planning full terminal credit while providing execution with positive terminal credit and teacher guidance. On all tested benchmarks, SCAD improves macro‑average accuracy by 4.48 percentage points on text tasks and 4.19 points on multimodal tasks compared to the strongest baseline.
By Shangyang Wu, Shuai Zhao, Ziyue Zhu, Jinyang Wu, Anh Tuan Luu, Haoran Luo
arXiv:2603. 19617v2 Announce Type: replace Abstract: Federated learning (FL) is a communication-efficient framework for solving distributed optimization problems.
By Mohammadjavad Ebrahimi, Daniel Burbano, Farzad Yousefian
The paper introduces XiangqiBench, an executable benchmark for evaluating large language model agents in Chinese chess. It tracks 8,568 multi‑turn trajectories from 12 frontier LLMs, revealing that metrics such as the Conversion Gap, Consistency Gap, and Simulation Gap overstate true closed‑loop success. The study shows that simply naming the correct move is insufficient; agents must reliably carry a plan through to a verified outcome against an opponent.
By Yekun Chai, Qiwei Peng, Haoyi Xiong
The paper investigates how large language model (LLM) agents exhibit source preference when selecting items in end‑to‑end search tasks across three domains. Across 12 agent models, each consistently favors certain sources and avoids others, even when items from disfavored sources better meet the user’s requirements. The study shows that revealing or hiding source identifiers influences selection, and that training rewards or missing information can create or mitigate these preferences.
By Jonghyun Song, Haewon Park, Jeonghoon Shim, Woojung Song, Yohan Jo
The paper "Code2Math: Can Your Code Agent Evolve Math Problems Through Exploration?" explores how code agents can autonomously generate more complex variations of existing math problems. It introduces a multi‑agent framework that evolves problems while ensuring they remain solvable and increasingly difficult. Experiments show that with sufficient exploration, code agents can produce new, structurally distinct, and challenging problems, demonstrating a scalable method for creating high‑difficulty mathematical reasoning tasks.
By Dadi Guo, Yuejin Xie, Qingyu Liu, Weixian Huang, Jiayu Liu, Zhiyuan Fan, Qihan Ren, Shuai Shao, Tianyi Zhou, Jianjie Feng, Wenze Su, Yujiu Yang, Dongrui Liu, Yi R. Fung
The paper introduces a simulation‑grounded vision‑language model (VLM) framework for wildfire monitoring that converts 2D wildfire simulations into labeled video episodes using a fixed Blender mapping to create low‑detail 3D proxies. These proxies, along with controllable video generation, provide a multimodal memory that a training‑free multi‑agent VLM system uses to retrieve reference episodes, reconcile visual and memory‑based predictions, and generate structured wildfire reports. The system achieves 77.3% accuracy on six simulator‑derived report fields, outperforming direct VLM querying and text‑only memory baselines.
By Duowen Chen, Yuchen Sun, Zhiqi Li, Yuxuan Liao, Sinan Wang, Bart van Bloemen Waanders, Bo Zhu
SymRegFlow is a symmetry‑regularized flow‑matching framework that enables multi‑view‑consistent video generation across continuously varying camera poses without requiring ground‑truth novel‑view RGB supervision. The method geometrically warps source views into noisy anchors and uses masked dual‑anchor supervision combined with cross‑anchor denoising‑output consistency to reduce anchor‑specific errors. Experiments on Cosmos‑Drive‑Dreams and nuScenes show that SymRegFlow achieves superior video quality, achieving the lowest FVD and FVMD scores and improving FID and instance preservation compared to existing baselines.
By Xi Ye, Yuzhu Wang, Xiaoyang Liu, Jiayi Wang, Yangyang Xu, Ruyu Wang, Wenlin Chen, Duo Su, Jun Zhu
PatchScene is a diffusion-based framework that completes large-scale LiDAR scenes using a patch-based voxel diffusion paradigm, generating fine-grained geometry in localized 3D regions. It incorporates a confidence-guided spatio-temporal fusion mechanism to merge overlapping patches and adjacent frames, and introduces an Annular-Flow diffusion strategy that propagates high-fidelity information from near to far ranges. Experiments on SemanticKITTI show state-of-the-art performance, with the model trained on 20 m LiDAR ranges generalizing to 50 m scenes without retraining.
By Qingdong Xu, Jiajun Zhu, Shilin Zhu, Xinjing He, Chao Lu, Huanran Wang, Jiyao Zhang
LiteTrajEval is a lightweight architecture designed for budget‑bounded trajectory evaluation of large‑language‑model agents. It creates compact domain‑specific rule profiles offline, preprocesses trajectories online to flag heuristic failure signals, and then uses a single rubric‑guided LLM judge to generate structured diagnostic reports. On public Magnetic‑One‑style and τ‑bench‑style datasets, LiteTrajEval improves failure‑localization alignment with human annotations by 20–35 percentage points and reduces cost and evaluation time by 6× and 8× respectively, and has been deployed in an enterprise agentic platform.
By Linh-An Phan, MingXue Wang, Guangyu Wu, Feng Pan, Zhaoyu Pang, Yanbin Zhang
The paper introduces the concept of any-start-time planning, addressing scenarios where the set of feasible states or actions changes over time, such as path planning among moving obstacles or train boarding. It presents a new data structure, the compound arrival time function (cATF), which compactly encodes the optimal plan as a function of start time and can be assembled via heuristic graph search. Experiments on the SIPP problem demonstrate that any-start-time algorithms using cATFs outperform traditional replanning approaches on difficult instances.
By Devin Wild Thomas (University of New Hampshire, USA), Solomon Eyal Shimony (Ben-Gurion University of the Negev, Israel), Wheeler Ruml (University of New Hampshire, USA), Erez Karpas (Technion - Israel Institute of Technology, Israel), Shahaf S. Shperberg (Ben-Gurion University of the Negev, Israel), Andrew Coles (King's College London, UK)
This thesis introduces methods for assessing and enhancing the robustness of large language models (LLMs) against adversarial input variations. It proposes a generative robustness metric, R_stab(f), based on Jensen‑Shannon divergence, and proves bounds for localized attacks. The work also presents adaptive attacks, Trojan detection techniques, defense strategies for multi‑layer systems, and lightweight attestation for Model Context Protocol (MCP) agentic systems, all implemented in the JudgeGuard, TrojanArmor, and MCPSec suites.
By Narek Maloyan
The paper introduces a joint design framework for mobile embodied AI networks (MEAN) that simultaneously optimizes agent movement, semantic compression, and transmit power to maximize energy efficiency. It formulates a non-convex max‑min EE problem and solves it using an alternating optimization–Dinkelbach algorithm, combining Dinkelbach transformation, successive convex approximation, and coordinate‑wise grid search. Simulation results demonstrate that this joint approach outperforms baselines that ignore mobility or compression, highlighting the advantages of integrating mobility control, semantic compression, and power allocation in MEAN.
By Yahao Ding, Jiaxiang Wang, Zhouxiang Zhao, Zhaohui Yang, Mingzhe Chen, Mohammad Shikh-Bahaei
The paper introduces FLINT, a domain‑specialized Text‑to‑SQL system designed for complex financial databases where data is stored as opaque integer keys. FLINT employs a lookup agent to resolve natural‑language concepts, uses embedding‑based retrieval of query templates from a compact expert‑authored bank, and prunes large schemas by traversing foreign‑key chains. Evaluated on 359 questions across two production financial schemas, FLINT outperforms state‑of‑the‑art baselines and is already deployed in a financial data retrieval service.
By Arijit Sehanobish, Bruno Gomes Coelho, Guillaume Michel, Sophia Zhi, Valerie Faucon-Morin, Kristen Howell
The paper introduces FinSkillBench, a benchmark of 2,603 financial AI agent episodes across portfolio construction, risk management, and fundamental analysis, evaluated with deterministic verifiers. Experiments on 9 models and 3 resource settings show that curated skill packages improve performance by +16.2 points, while in‑episode skill generation adds only +0.5 points. Decomposing the premium reveals that executable tools contribute +19.5 points, documentation +5.6 points, and their combination is subadditive, with the effect varying by workflow type.
By Jermyn Zhen Yong Bek, Zhuang Qiang Bok, Zhongtian Sun
HazardWeaver is an AI agent designed to select and execute appropriate scientific methods for hazard analysis by leveraging a Hazard Knowledge Compiler and a Hazard Capability Graph. It dynamically revises its route choices as new evidence and execution results emerge, ensuring that selected methods are both applicable and executable with available data and tools. The system is evaluated on the Hazard Weaver Benchmark, which includes 141 instances across multiple hazard domains, and demonstrates superior performance over existing agent systems, especially when multiple scientific routes are viable.
By Wangshu Zhu, Xueqi Cheng, Liang Wu, Yushun Dong