Proposes daVinci-Dev, using agent-native mid-training with context and environment trajectories to enhance autonomous software engineering capabilities.

Ji Zeng, Dayuan Fu, Tiantian Mi et al.

TriPlay-RL uses tri-role self-play reinforcement learning for LLM safety alignment, improving adversarial effectiveness by 20%-50%.

Zhewen Tan, Wenhan Yu, Jianfeng Si et al.

TEFormer enhances spiking Transformers with bidirectional temporal fusion, achieving superior performance across datasets.

Sicheng Shen, Mingyang Lv, Bing Han et al.

Using polytope decomposition, enhance ReLU network approximation, especially near singular points.

ZeYu Li, ShiJun Zhang, TieYong Zeng et al.

Yunjue Agent achieves zero-start adaptability for open tasks via parallel batch evolution and tool optimization, outperforming baselines.

Haotian Li, Shijun Yang, Weizhen Qi et al.

GAIA system enhances GUI test-time scaling performance using a data flywheel, significantly improving action accuracy.

Shaokang Wang, Pei Fu, Ruoceng Zhang et al.

Proposes ATLAS, a latent-space transformer framework combining diffusion, stochastic interpolants, and CRPS for state-of-the-art probabilistic medium-range weather forecasting, outperforming ECMWF IFS.

Jean Kossaifi, Nikola Kovachki, Morteza Mardani et al.

UniCog analyzes LLM cognitive abilities via latent mind space, improving reasoning performance by 7.5%.

Jiayu Liu, Yinhe Long, Zhenya Huang et al.

SRA 2 accelerates diffusion model training using VAE feature alignment, adding only 4% GFLOPs.

Mengmeng Wang, Dengyang Jiang, Liuzhuozheng Li et al.

CoT-Seg leverages chain-of-thought reasoning and self-correction, enabling training-free, robust segmentation with 10%+ improvements on challenging datasets.

Shiu-hong Kao, Chak Ho Huang, Huaiqian Liu et al.

Proposes PPIA, a black-box physical prompt injection attack using visual observation with 98% success, exploiting environment cues.

Chen Ling, Kai Hu, Hangcheng Liu et al.

Proposes a three-dimensional query performance prediction (QPP) evaluation framework (SRMQ-PP, MRSQ-PP, MRMQ-PP), demonstrating significant performance differences across models and tasks, with traditional score-based models excelling in multi-model scenarios.

Payel Santra, Partha Basuchowdhuri, Debasis Ganguly

Introduces Reddit-Amazon-EM dataset; evaluates SOTA entity matching methods; GNEM achieves 96.29% F1.

Zihan Huang, Rohan Surana, Zhouhang Xie et al.

Introduced TAC and Soft-TAC methods, significantly enhancing reward function design efficiency and accuracy.

Calarina Muslimani, Yunshu Du, Kenta Kawamoto et al.

RPEval benchmark reveals personalized memory's impact on LLM behavior; RP-Reasoner significantly reduces irrational personalization.

Xueyang Feng, Weinan Gan, Xu Chen et al.

This study evaluates LLM-simulated users in τ-Bench, revealing robustness, validity, and fairness issues with success rate differences up to 9 percentage points across models.

Preethi Seshadri, Samuel Cahyawijaya, Ayomide Odumakinde et al.

CORD employs weighted on-policy cross-modal self-distillation, using text as internal teacher, to reduce audio-text reasoning gap by over 40%.

Jing Hu, Danxiang Zhu, Xianlong Luo et al.

Analyzes gradient descent for shallow Transformers, finding width scales logarithmically with sample size, optimization error independent of sequence length.

Enes Arda, Semih Cayci, Atilla Eryilmaz

Deep binding of virtual personas with rich backgrounds and time grounding enhances LLMs' accuracy in social behavior simulation, matching human data closely.

Suhong Moon, Minwoo Kang, Joseph Suh et al.

Proposes DSGym, a modular framework for end-to-end evaluation and training of data science agents, covering diverse scientific and predictive tasks.

Fan Nie, Junlin Wang, Harper Hua et al.