Researcher — paper.dinq.me

Multimodal Large Language Models (MLLMs) frequently suffer from unfaithfulness, generating reasoning chains that drift from visual evidence or contradict final predictions. We propose Faithful-First Reasoning, Planning, and Acting (RPA) framework in which FaithEvi provides step-wise and chain-lev...

多模态推理评测与基准OpenReview 项目页

作者位次 #4/9ACL 2026Findings· Findings of ACL

S tream M e C o: Long-Term Agent Memory Compression for Efficient Streaming Video Understanding

JunXi Wang · Te Sun · Jiayi Zhu · Junxian Li · Haowen Xu · Zichen Wen · Xuming Hu · Zhiyu li · Linfeng Zhang

摘要

Vision agent memory has shown remarkable effectiveness in long-video understanding; however, storing such memory for videos incurs substantial overhead, leading to high costs in both storage and computation. To address this issue, we propose StreamMeCo, an efficient Stream Agent Memory Compressio...

效率与系统多模态检索与知识OpenReview 项目页

作者位次 #2/9KDD 2026Oral· Knowledge Discovery & Data Mining

Speak-to-Structure: Evaluating LLMs in Open-domain Natural Language-Driven Molecule Generation

Jiatong LI · Junxian Li · Weida Wang · Yunqing LIU · Changmeng Zheng · D institute China-Beijing) · Dongzhan Zhou · Xiao-Yong Wei · Li Qing

评测与基准检索与知识OpenReview

Junxian Li

获取完整研究档案

联系方式

2026 顶会论文

PermuQuant: Lowering Per-Group Quantization Error by Reordering Channels for Diffusion Models

AttnDiff: Attention-based Differential Fingerprinting for Large Language Models

TwinPrune: Density-Aware Two-Phase Token Pruning for Vision-Language Models

IAG: Input-aware Backdoor Attack on VLM-based Visual Grounding

S$^2$-MLLM: Boosting Spatial Reasoning Capability of MLLMs for 3D Visual Grounding with Structural Guidance

Faithful-First Reasoning, Planning, and Acting for Multimodal LLM s

S tream M e C o: Long-Term Agent Memory Compression for Efficient Streaming Video Understanding

Speak-to-Structure: Evaluating LLMs in Open-domain Natural Language-Driven Molecule Generation

常见合作者