The paper introduces Temporal Semantic Memory (TSM), a framework that improves how large language model agents manage memory by addressing two key shortcomings: temporal inaccuracy and temporal fragmentation. TSM constructs a semantic timeline instead of a dialogue timeline, consolidating temporally continuous and semantically related information into durative memory. During retrieval, it aligns the query’s temporal intent with the semantic timeline, enabling the use of temporally appropriate durative memories and yielding up to a 12.2% accuracy boost over existing methods.
By Miao Su, Yucan Guo, Zhongni Hou, Long Bai, Zixuan Li, Yufei Zhang, Guojun Yin, Wei Lin, Xiaolong Jin, Jiafeng Guo, Xueqi Cheng
UniToolCall introduces a unified framework for tool-use in large language model agents, standardizing toolset construction, dataset generation, and evaluation. The framework aggregates over 22,000 tools and creates a hybrid training corpus of more than 390,000 instances by combining ten public datasets with synthetically generated, structurally controlled trajectories. It models diverse interaction patterns—single‑hop vs. multi‑hop, single‑turn vs. multi‑turn, serial vs. parallel execution—and adds an Anchor Linkage mechanism to enforce cross‑turn dependencies, while converting seven public benchmarks into a common Query–Action–Observation–Answer format for fine‑grained evaluation.
By Yijuan Liang, Xinghao Chen, Yifan Ge, Ziyi Wu, Hao Wu, Changyu Zeng, Wei Xing, Xiaoyu Shen
OptimusKG is a multimodal biomedical labeled property graph that integrates structured and semi‑structured resources to preserve detailed, type‑specific metadata across molecular, anatomical, clinical, and environmental domains. The graph contains nearly 191,000 nodes, over 21.8 million edges, and more than 67 million property instances derived from 18 ontologies, with a top‑level schema that enforces node and edge constraints while retaining granular provenance. Validation using the PaperQA3 agent found that 70.0% of sampled edges are supported by literature evidence, and the graph offers a standardized resource for machine learning, knowledge‑grounded retrieval, and hypothesis generation in biomedical research.
By Lucas Vittor, Ayush Noori, I\~naki Arango, Joaqu\'in Polonuer, Sam Rodriques, Andrew White, David A. Clifton, Marinka Zitnik
The paper introduces K‑GAT, a neuro‑symbolic framework that generates multi‑agent collaboration topologies conditioned on external evidence, treating the design as a knowledge‑conditioned structure learning problem. Unlike prior methods that rely mainly on large language model parameters, K‑GAT integrates external evidence directly into autoregressive graph generation, reducing redundant interactions and improving verification in knowledge‑intensive tasks. Experiments on benchmarks such as the expert‑level GPQA dataset show K‑GAT achieving a +15.7% accuracy gain over the LLM‑Debate baseline while using fewer computational tokens.
By Yangxiao Jiang, Jiarun Fan, Mingcong Xu, Yanxi Guo, Jiwen Feng, Shanqing Xu, Mengchen Qian, Wei Chen, Xiaojin Zhang
The paper introduces a method that integrates action abstraction into policy optimization for reinforcement learning and generative flow networks. By iteratively identifying frequently used action subsequences in high‑reward trajectories and treating them as single high‑level actions, the approach expands the action space and improves sample efficiency. Experiments on synthetic and real‑world tasks show that this technique discovers diverse high‑reward states more effectively, especially on challenging exploration problems, and yields interpretable abstract actions that reflect the underlying reward structure.
By Oussama Boussif, L\'ena N\'ehale Ezzine, Joseph D Viviano, Micha{\l} Koziarski, Moksh Jain, Esmeralda S. Whitammer, Emmanuel Bengio, Rim Assouel, Yoshua Bengio
Multimodal Language Models as Text-to-Image Model Evaluators presents MT2IE, a framework where a multimodal large language model generates evaluation prompts and scores images, achieving higher correlation with human judgment than prior metrics. MT2IE recovers official T2I model rankings using only 20 prompts—far fewer than traditional benchmarks—and adapts prompts to each model’s performance, maintaining informative scoring ranges. The approach demonstrates that dynamic, interactive evaluation can replace static benchmarks as T2I models improve.
By Jiahui Chen, Candace Ross, Reyhane Askari-Hemmat, Koustuv Sinha, Melissa Hall, Amy Zhang, Michal Drozdzal, Adriana Romero-Soriano
The paper proposes a single variational principle that explains how gating mechanisms, their dynamics, and neural implementations for behavioral composition can be unified. This principle yields softmax gating, an energy‑based dynamical system with guaranteed convergence, and a recurrent neural network model with context‑dependent, local interactions. Experiments across collective behavior, human decision‑making, and layered control show that the mechanism reproduces known behavioral patterns, offers interpretable accounts of behavior combination, and matches or outperforms existing methods.
By Francesca Rossi, Veronica Centorrino, Francesco Bullo, Giovanni Russo
The paper introduces AutoMat, a benchmark designed to test large language model (LLM) coding agents on their ability to reproduce claims from computational materials science. AutoMat presents three challenges: reconstructing underspecified procedures, navigating specialized toolchains, and assessing whether the evidence supports a claim. Experiments show that current LLM agents achieve low success rates, with the best setting reaching only 53%, and failures stem mainly from incomplete procedures, methodological deviations, and execution fragility.
By Ziyang Huang, Yi Cao, Ali K. Shargh, Jing Luo, Ruidong Mei, Mohd Zaki, Zhan Liu, Wyatt Bunstine, William Jurayj, Somdatta Goswami, Tyrel McQueen, Michael Shields, Jaafar El-Awady, Paulette Clancy, Benjamin Van Durme, Nicholas Andrews, William Walden, Daniel Khashabi
AdaMem introduces adaptive memory policies that allow personalized agents to decide what information to write into long‑term memory based on user preferences for each interaction context. Each policy is updated from periodic feedback and controls subsequent memory writing, aiming to improve relevance and reduce unnecessary memory persistence. In experiments on AdaMem‑Bench, AdaMem raises QA accuracy from 80.0% to 84.35% while cutting persistent memory by 9.27%, though models still struggle to execute policies reliably.
By Xingyu Chen, Rui Wang, Zhaopeng Tu, Liefeng Bo
SpecMine is a large-scale corpus that documents Spec-Driven Development (SDD) artifacts in public GitHub repositories. It includes a broad census of 470,795 spec files from 73,030 repositories linked to 17 tools, a focused census of 98,574 Kiro layout files from 12,910 repositories, and a sweep of 5,992 pull requests across 581 repositories that modify specs. The dataset provides enriched metadata, full commit histories, parsed document structures, and over 2.4 million typed references connecting specs to code, sibling documents, PRs, branches, and issues.
By Shyam Agarwal, Anmol Singhal, Travis Breaux, Bogdan Vasilescu
The paper introduces InternGeometry, a large language model agent that achieves medalist-level performance on International Mathematical Olympiad geometry problems. It overcomes traditional heuristic limitations by iteratively proposing and verifying auxiliary constructions with a symbolic engine, supported by a dynamic memory mechanism that allows over 200 interactions per problem. Using Complexity-Boosting Reinforcement Learning, InternGeometry trains on only 13,000 examples—0.004% of the data used by AlphaGeometry 2—and solves 44 of 50 IMO geometry problems, surpassing the average gold medalist score.
By Haiteng Zhao, Junhao Shen, Yiming Zhang, Songyang Gao, Kuikun Liu, Tianyou Ma, Fan Zheng, Dahua Lin, Wenwei Zhang, Kai Chen
RILA is an execution‑driven agent that integrates browser rendering into the generation loop for interactive web development. It uses an Action Interaction Verification module to replay reference interactions on generated pages, collecting execution‑aware observations, and an Execution‑aware Rendering Score to jointly assess interaction correctness and visual fidelity during iterative optimization. A data synthesis pipeline further augments training data, enabling RILA to significantly improve interaction and visual quality across foundation models, even outperforming larger one‑shot generators.
By Yilong Guo, Hanqi Chen, Zixiao Ye, Guanzhong Wang, Chen Yu, Zeyu Chen
VIPS is a benchmark for vehicle‑to‑infrastructure cooperative autonomous driving that uses pseudo‑simulation to combine vehicle and infrastructure observations, enabling scalable yet realistic evaluation of robustness and error propagation without full simulation. The paper also introduces CoS‑V2X, a cooperative planning framework that employs sparse representations to model vehicle‑infrastructure interactions efficiently and robustly under heterogeneous observations.
By Hoonhee Cho, Jae-Young Kang, Giwon Lee, Hyemin Yang, Heejun Park, Kuk-Jin Yoon
BOLT is a lightweight plug‑and‑play module that enables preparation‑free heterogeneous cooperative perception by adapting neighboring features online through ego‑as‑teacher distillation. It requires only ego predictions, no ground‑truth labels, and uses high‑confidence ego features to align cross‑agent feature domains while allowing neighbors to contribute in low‑confidence regions. With just 0.9 M trainable parameters, BOLT boosts AP@50 by up to 32.3 points over unadapted fusion and consistently outperforms ego‑only results on DAIR‑V2X and OPV2V.
By Kang Yang, Tianci Bu, Peng Wang, Deying Li, Yongcai Wang
The thesis "Robotic Contextual Awareness for Human-Robot Collaboration and Environmental Understanding" addresses the challenge of enabling autonomous mobile robots to operate safely in dynamic, human-centric environments by developing advanced contextual awareness. It presents two complementary research directions: (1) a human re-identification and tracking system that allows a robot to recognize and collaborate with a specific person while ignoring others, and (2) enhanced perceptual capabilities that provide geometric and semantic understanding of the environment for improved motion planning and interaction. These contributions aim to improve robots’ knowledge of their surroundings, facilitating smoother and more natural collaboration with humans.
By Federico Rollo
The paper introduces the concept of an epistemic Sybil problem in multi‑agent AI systems, where multiple agents may produce seemingly independent reports that actually stem from the same underlying evidence. It formalizes this issue using information‑theoretic measures and demonstrates through large‑scale experiments that naive aggregation of replicated reports can severely degrade inference accuracy unless the system accounts for shared evidence ancestry and correlated extraction errors. The study shows that aggregators that track evidential dependence rather than merely report multiplicity or similarity achieve better calibration and inference performance.
By Marc Bara
The paper introduces ClaimReceipt, a specification and verifier that checks whether a claim in an agent evaluation can be recomputed from retained evidence (sufficiency) and whether the evidence covers the entire experiment set (coverage). Using the CR‑2 verifier on 1,392 historical records, the authors demonstrate accurate reproduction of audit verdicts, non‑redundant field groups, and zero false positives on semantic faults. In a prospective CR‑3 run, the system correctly flags missing receipts and preserves coverage when private evidence is withheld, while adding minimal overhead to inference time and transaction size.
By Peiying Zhu, Sidi Chang
The paper introduces Codebook Agent, a method for adapting the communication topology of large language model (LLM) multi‑agent systems to individual queries. Instead of searching the full adjacency space, it compresses successful topologies into a 16‑entry codebook via a vector‑quantized autoencoder and uses a reward‑weighted MLP to select a code for each query, followed by a single‑pass MLP proxy to rerank candidates. This approach eliminates iterative search and message passing at test time, achieving higher accuracy across six benchmarks, faster topology generation (2.4 ms), and reduced token usage (21.9–33.2 %).
By Jinxi Yu, Yubei Li, Eric Hanchen Jiang, Zhi Zhang, Dong Liu, Wenxiao Zhao, Levina Li, Kai-Wei Chang, Ying Nian Wu
The paper introduces SCX Router, a lightweight GLiClass-based model selector that assigns suitability scores to inference-time language models without autoregressive generation. It uses a 0.6B-parameter Qwen3 decoder with a shallow bidirectional scorer, preserving a text-only key–value cache across sessions and predicting task attributes such as type, difficulty, and expected output length. The authors build a comprehensive task ontology with 23 families, 115 types, and 1,173 synthetic examples, generating 150,000 verifier-scored tasks to train the router, which outperforms baseline models on LiveBench subsets with a top‑1 score of 0.707 versus 0.696 for the strongest fixed model.
By Ihor Stepanov, Aleksandr Smechov, Mykhailo Shtopko, Dmytro Vodianytskyi, Oleksandr Lukashov
Hybrid Retrieval-Augmented Generation with Knowledge Graph Expansion, RRF Fusion, and Per-Chunk Grounded Evaluation for Enterprise Document Search describes DocuSearch, an offline multi‑agent system designed for telecom network operations. The system combines semantic vector search, BM25 full‑text search, and knowledge‑graph neighbor expansion, merges the results via Reciprocal Rank Fusion, and reranks with a cross‑encoder before pruning with Maximal Marginal Relevance. A per‑chunk evaluation loop ensures only grounded answers are returned, achieving Precision@10 of 0.69, Recall@10 of 0.79, and an 89.6% grounding rate—improvements of 15, 16, and 18.4 percentage points over a dense‑only baseline.
By Harish Saragadam, Sudhanshu Sharma, Meghana Pujari