Researcher — paper.dinq.me

2026 顶会论文

作者位次 #3/7ICML 2026Poster· General Machine Learning / Hardware and Software

HARD-KV: Head-Adaptive Regularization for Decoding-time KV Compression

Yuxuan Yang · Feiyang Ren · Bowen Zeng · Dalin Zhang · Jinpeng Chen · Gang Chen · Huan Li

代码效率与系统OpenReview

一作ACL 2026Poster· Natural Language Processing

H ybrid KV : Hybrid KV Cache Compression for Efficient Multimodal Large Language Model Inference

Bowen Zeng · Feiyang Ren · Jun Zhang · Xiaoling Gu · Ke Chen · Lidan Shou · Huan Li

摘要

Multimodal Large Language Models (MLLMs) have advanced unified reasoning over text, images, and videos, but their inference is hindered by the rapid growth of key–value (KV) caches. Each visual input expands into thousands of tokens, causing caches to scale linearly with context length and remain...

效率与系统多模态检索与知识OpenReview 项目页

作者位次 #5/10ACL 2026Findings· Findings of ACL

Efficient Inference for Large Vision-Language Models: Bottlenecks, Techniques, and Prospects

Jun Zhang · Yicheng Ji · Feiyang Ren · Yihang Li · Bowen Zeng · Zonghao Chen · Ke Chen · Lidan Shou · Gang Chen · Huan Li

摘要

Large Vision-Language Models (LVLMs) enable sophisticated reasoning over images and videos, yet their inference is hindered by a systemic efficiency barrier known as visual token dominance. This overhead is driven by a multi-regime interplay between high-resolution feature extraction, quadratic a...

多模态效率与系统代码OpenReview 项目页

Bowen Zeng

获取完整研究档案

联系方式

2026 顶会论文

HARD-KV: Head-Adaptive Regularization for Decoding-time KV Compression

H ybrid KV : Hybrid KV Cache Compression for Efficient Multimodal Large Language Model Inference

Efficient Inference for Large Vision-Language Models: Bottlenecks, Techniques, and Prospects

常见合作者