This study evaluates multilingual keyword and content variant injection attacks on LLM relevance judgments, revealing persistent inflation and evasion of defenses.

Nguyen Khoi Vo, Duy Duong Tuong, Oleg Zendel et al.

Proposes Who&When Pro, using error injection to generate 12,326 high-quality failure traces across multimodal tasks for automated failure attribution.

Jiale Liu, Huajun Xi, Shaokun Zhang et al.

A ReAct retrieval and fine-tuned ensemble reaches 0.151 class-balanced Brier on 404 M&A deals, outperforming markets, XGBoost, and frontier LLMs.

Hinal Jajal, Michal Mucha, Charles Sweat et al.

DenseAR achieves multimodal autoregressive visual modeling using a single-scale tokenizer, enhancing ImageNet generation quality.

Chicago Y. Park, Jialin Mao, Xiaojian Xu et al.

MOSAIC combines agent-centric semantic memory and ILP optimization to enhance multi-agent embodied planning, achieving 30% faster execution and fewer failures.

Kunjal Panchal, Saayan Mitra, Sunav Choudhary et al.

SigLIP-HD employs fine-to-coarse supervision to enhance visual perception at standard resolution, outperforming baseline models with no extra inference cost.

Lihe Yang, Zhen Zhao, Hengshuang Zhao

Test-Time Scaling (TTS) significantly improves reasoning in small VLMs on multilingual visual MCQ, achieving 84.1% accuracy in the best configuration.

Spiros Baxevanakis, Peng-Jian Yang

Mach-Mind-4-Flash, a 35B MoE model with 3B activated parameters, matches 100B-class performance.

Foundation Model Team

Proposes a dynamic inverse rendering framework combining object tracking and surface reconstruction, significantly improving material-light disentanglement.

Raza Yunus, Benjamin Ummenhofer, Jan Eric Lenssen et al.

Proposes risk-aware GUMDP framework using ERM and MCTS, enabling robust multi-task decision-making under uncertainty.

Pedro P. Santos, Fábio Vital, Alberto Sardinha et al.

Super method uses Wanda score for sparse fine-tuning, excelling on Math17K dataset.

Ivan Ilin, Philip Zmushko, Peter Richtárik

Proposes a blockchain-linked, auditable request management framework integrating ML, LLM, and blockchain for telecom/IoT fraud control, enabling request-level policy and traceability.

Saviz Changizi, Nasibeh Mohammadzadeh, Mohammad Shojafar et al.

MedRealMM evaluates multimodal medical consultations using MCCP framework, featuring 5,620 real cases.

Runhan Shi, Quan Zhou, Yuqian Xu et al.

OmniMapBench uses VDI to validate visual reasoning on maps; top LVLM achieves only 75.03% accuracy.

Yang Chen, Yunwen Li, Yufan Shen et al.

MOSAIC employs multi-objective MIP to automatically search heterogeneous Transformer architectures, combining linear, sparse, and low-rank mechanisms, boosting efficiency by over 2.5×.

Yuncheng Yang, Feiyang Ye, Shixian Luo et al.

Proposes CABS-C and CABS-D algorithms leveraging surrogate rewards and correlations to improve LLM routing efficiency and reduce regret.

Ajay Narayanan Sridhar, Ronak Singh, Mehrdad Mahdavi et al.

Proposes SATS and Token Routing to improve activation sparsification in LLMs, enhancing efficiency and performance.

Bishmoy Paul, Youngmin Yi, Hoeseok Yang

MultiView-Bench evaluates VLM multi-view spatial reasoning; models struggle with 3D relations, improved by active viewpoint fusion (+12-20%).

Hantao Zhang, Jinru Sui, Ed Li et al.

Proposes a novel language model fine-tuning framework using pointwise constraints, significantly reducing tail constraint violations.

Ignacio Hounie, Ignacio Boero, Alejandro Ribeiro

Introduces Long-Horizon-Terminal-Bench with 46 long-horizon tasks, emphasizing dense rewards and partial credit, to evaluate models' sustained performance.

Zongxia Li, Zhongzhi Li, Yucheng Shi et al.