Proposes Predictive Regularization (PRe) to mitigate visual representation degradation in Multimodal Large Language Models, boosting visual fidelity and task performance.

Enguang Wang, Qiang Wang, Yuanchen Wu et al.

Proposed TARNet for bias-robust uplift modeling; validated via semi-synthetic data under structural biases.

Yuxuan Yang, Dugang Liu, Yiyan Huang

LumosX uses relational self-attention and cross-attention for personalized video generation, enhancing face-attribute alignment.

Jiazheng Xing, Fei Du, Hangjie Yuan et al.

VideoSeek actively seeks critical evidence using video logic flow, reducing frame usage by 93% and improving LVBench accuracy by 10.2 points.

Jingyang Lin, Jialian Wu, Jiang Liu et al.

SALM framework improves generalization in cybersecurity tasks using multi-modal contrastive learning, especially for text and payload data.

Jianan Huang, Rodolfo V. Valentim, Luca Vassio et al.

Semantic Token Clustering (STC) method achieves efficient uncertainty quantification in large language models, significantly reducing computational overhead.

Qi Cao, Andrew Gambardella, Takeshi Kojima et al.

Study of SFT-DPO interaction in small models reveals full fine-tuning outperforms LoRA.

Yuming Feng, Christy Yang

Proposes a Lie algebra-based homomorphic latent dynamics model within JEPA for wireless CSI prediction and spatial representation.

Salmane Naoumi, Mehdi Bennis, Marwa Chafii

IsoCLIP decomposes CLIP projectors for efficient intra-modal alignment, enhancing retrieval performance.

Simone Magistri, Dipam Goswami, Marco Mistretta et al.

This study employs Lipschitz theory to analyze how demonstrations, CoT, and prompts influence ICL performance.

Xuhan Tong, Yuchen Zeng, Jiawei Zhang

LoD-Loc v3 employs instance silhouette alignment with synthetic data to improve urban aerial localization, achieving over 97% success in 2m/2° accuracy.

Shuaibang Peng, Juelin Zhu, Xia Li et al.

Proposes Recoding-Decoding (RD) to enhance LLMs' sustained creativity and diversity, surpassing modal decoding limits.

Queenie Luo, Gary King, Michael Puett et al.

VEGA-3D leverages implicit 3D priors in video generation models to enhance scene understanding.

Xianjin Wu, Dingkang Liang, Tianrui Feng et al.

Matryoshka Gaussian Splatting (MGS) enables continuous level of detail control without sacrificing full-capacity rendering quality.

Zhilin Guo, Boqiao Zhang, Hakan Aktas et al.

Vision pathways dominate action generation in VLA models; language sensitivity is task-dependent.

Bryce Grant, Xijia Zhao, Peng Wang

MonoArt uses progressive structural reasoning for monocular 3D reconstruction, achieving improved accuracy and speed on the PartNet-Mobility dataset.

Haitian Li, Haozhe Xie, Junxiang Xu et al.

NavTrust benchmarks embodied navigation robustness by systematically introducing RGB, depth, and instruction corruptions, revealing significant robustness gaps in current models.

Huaide Jiang, Yash Chaudhary, Yuping Wang et al.

MoTok method reduces trajectory error from 0.72 cm to 0.08 cm and FID from 0.083 to 0.029 on HumanML3D.

Chenyang Gu, Mingyuan Zhang, Haozhe Xie et al.

SAMA achieves instruction-guided video editing through semantic anchoring and motion alignment, significantly enhancing editing precision and motion consistency.

Xinyao Zhang, Wenkai Dong, Yuxin Song et al.

EffectErase uses reciprocal learning for high-quality video object removal and insertion, leveraging the VOR dataset.

Yang Fu, Yike Zheng, Ziyun Dai et al.