Proposed BiLaLoRA with CLIP-based H2C loss achieves superior real-world image dehazing with efficient parameter tuning.

Yan Zhang, Long Ma, Yuxin Feng et al.

LLM-assisted MIPVU rule script generation enables interpretable Chinese metaphor identification; protocol choice is the main source of variation.

Weihang Huang, Mengna Liu

RAGPerf is an end-to-end benchmarking framework for retrieval-augmented generation systems, supporting various datasets and embedding models with negligible performance overhead.

Shaobo Li, Yirui Zhou, Yuan Xu et al.

PrefixWall monitors cache reuse and selectively isolates prefixes to prevent APC side channels, boosting performance by 70%.

Panagiotis Georgios Pennas, Konstantinos Papaioannou, Marco Guarnieri et al.

Using structured linked data as a memory layer improves RAG system retrieval accuracy by 29.6% in standard RAG and 29.8% in agentic pipeline.

Andrea Volpini, Elie Raad, Beatrice Gamba et al.

SCORE replaces layer stacking with recurrent shared blocks inspired by ODEs, improving training speed and reducing parameters in GNNs, MLPs, and Transformers.

Guillaume Godin

This paper presents an event-driven E-Skin system with dynamic binary scanning and real-time SNN classification, achieving a 12.8x scan reduction and 92.11% accuracy.

Gaishan Li, Zhengnan Fu, Anubhab Tripathi et al.

V2A-DPO enhances video-to-audio generation through Direct Preference Optimization, significantly improving semantic consistency and temporal alignment.

Nolan Chan, Timmy Gang, Yongqian Wang et al.

Proposes World2Act, a latent-space post-training framework that improves success rates by +2.5% using contrastive alignment without pixel supervision.

An Dinh Vuong, Tuan Van Vo, Abdullah Sohail et al.

Graph-GRPO introduces analytical transition probabilities and reinforcement learning for efficient graph flow model training, achieving 95% validity in 50 steps.

Baoheng Zhu, Deyu Bo, Delvin Ce Zhang et al.

CRC-CBF provides risk-adaptive probabilistic safety for human-robot navigation, but the supplied paper text reports no numerical outcome metrics.

Jake Gonzales, Kazuki Mizuta, Karen Leung et al.

ReST-RL combines pre-trained locomotion policies with residual modules, achieving 96.9% success in stable humanoid tray transport under disturbances.

Anlun Huang, Zhenyu Wu, Soofiyan Atar et al.

A memory–retrieval–reflection agent study with 51 users and 4 concepts found distributional calibration but weak identity fidelity.

Zichao Wang, Alexa Siu

TATIC uses torque estimation and TCN to achieve a 0.904 Macro-F1 score in intent recognition.

Jiurun Song, Xiao Liang, Minghui Zheng

MIPO enhances LLM performance by maximizing mutual information between prompts and responses without additional data.

Hyunji Nam, Haoran Li, Natasha Jaques

OpenClaw-RL employs an asynchronous architecture to leverage next-state signals for online personalized and general agent training, integrating evaluative and directive signals.

Yinjie Wang, Xuyang Chen, Xiaolong Jin et al.

Introduced DOWIS dataset to evaluate SLLMs in multilingual settings, finding text prompts outperform spoken prompts.

Maike Züfle, Sara Papi, Fabian Retkowski et al.

InternVL-U is a 4B-parameter multimodal model achieving state-of-the-art understanding, reasoning, generation, and editing, outperforming larger models.

Changyao Tian, Danni Yang, Guanzhou Chen et al.

N-gram models predict reading time best due to sensitivity to simple statistics.

James A. Michaelov, Roger P. Levy

Proposes Step-Aware Contrastive Alignment (SACA), leveraging step-by-step evaluation to improve vision-language navigation in continuous environments.

Haoyuan Li, Rui Liu, Hehe Fan et al.