OVERTONE method mitigates heterogeneous token overfitting in LLM knowledge editing by adaptively smoothing the target distribution.

Tianci Liu, Ruirui Li, Zihan Dong et al.

FastVim halves Mamba scan depth through alternating spatial pooling, achieving up to 72.5% faster inference on 2048² images.

Saarthak Kapse, Robin Betz, Srinivasan Sivanandan

M+ extends MemoryLLM with a long-term memory mechanism, enabling over 160k tokens retention, significantly improving long-text understanding.

Yu Wang, Dmitry Krotov, Yuanzhe Hu et al.

OrcaLoca enhances software issue localization accuracy with priority scheduling and context pruning, achieving a 65.33% function match rate.

Zhongming Yu, Hejia Zhang, Yujie Zhao et al.

STP employs self-play with iterative conjecturing and proving, achieving 28.5% proof rate, doubling prior best.

Kefan Dong, Tengyu Ma

Planalyst tool enables quantitative reasoning on plan spaces, supporting complex reasoning modes.

David Speck, Markus Hecher, Daniel Gnad et al.

Proposes a carefree multiple testing framework using e-process supremum and adjusters to control FDR under arbitrary dependence.

Yury Tavyrikov, Jelle J. Goeman, Rianne de Heide

Reward-Guided Speculative Decoding (RSD) combines lightweight draft models with reward signals to improve LLM inference efficiency, reducing up to 4.4× FLOPs while boosting accuracy.

Baohao Liao, Yuhui Xu, Hanze Dong et al.

Introduces the GO multimodal dataset with six sensors, supporting perception and navigation in unstructured outdoor environments.

Peng Jiang, Kasi Viswanath, Akhil Nagariya et al.

EgoMe dataset with 7902 pairs, multimodal data, enhances robot imitation via cross-view alignment.

Heqian Qiu, Zhaofeng Shi, Lanxiao Wang et al.

Proposed a dynamic model identification-based gravity compensation for dVRK-Si PSM, constructing a full kinematic model with improved control accuracy.

Haoying Zhou, Hao Yang, Anton Deguet et al.

OmniPhysGS employs learnable constitutive Gaussians for multi-material 3D dynamic scene synthesis, achieving 3-16% better visual and text alignment metrics.

Yuchen Lin, Chenguo Lin, Jianjin Xu et al.

Large-scale controlled study shows data contamination inflates BLEU scores up to 30 points, especially in 8B models, with contamination timing and format affecting results.

Muhammed Yusuf Kocyigit, Eleftheria Briakou, Daniel Deutsch et al.

SAM2Act integrates large-scale visual foundation models with multi-view transformers, achieving 86.8% success and enhanced spatial memory.

Haoquan Fang, Markus Grotz, Wilbert Pumacay et al.

This study analyzes the role of feed-forward layers in Transformer-based in-context nonlinear learning, proposing a GLU-based model for polynomial functions.

Haoyuan Sun, Ali Jadbabaie, Navid Azizan

Janus-Pro scales up to 7B parameters, employing optimized training, expanded data, and decoupled visual encoding, achieving state-of-the-art multimodal understanding and text-to-image generation.

Xiaokang Chen, Zhiyu Wu, Xingchao Liu et al.

Proposes improved second-order methods (NALEN, CALEN) with gradient complexities O(¯d+¯d^{1/3}ε^{-3/2}) and O((¯d+¯d^{13/21}ε^{-2/7})ln¯d), surpassing prior results.

Lesi Chen, Chengchang Liu, Luo Luo et al.

This study compares SFT and RL in foundation models, showing RL achieves 77.8% success in OOD generalization, outperforming SFT by 33.8%.

Tianzhe Chu, Yuexiang Zhai, Jihan Yang et al.

LLM-AutoDiff introduces a graph-based automatic prompt optimization framework using textual gradients, supporting multi-component and cyclic LLM workflows, significantly improving accuracy and efficiency.

Li Yin, Zhangyang Wang

iRe-VLA alternates frozen-backbone online RL with full-model imitation, raising MetaWorld unseen-task success to 0.80.

Yanjiang Guo, Jianke Zhang, Xiaoyu Chen et al.