Xin HE's Website

GM-Skip: 基于度量引导的 Transformer 块跳过策略加速视觉语言模型 | Metric-Guided Transformer Block Skipping for Efficient VLMs

2 min read   ·   March 12, 2026

2026   ·   VLM   Transformer   Efficiency   Inference-Acceleration     ·   research  

Ghost in the Cloud: 地理分布式大模型训练的安全隐患 | Your Geo-distributed LLM Training is Easily Manipulated

2 min read   ·   March 12, 2026

2026   ·   LLM   Security   Federated-Learning   Distributed-Training   ICLR     ·   research  

Lang-PINN: 从自然语言到物理信息神经网络的多智能体框架 | From Language to PINNs via a Multi-Agent Framework

2 min read   ·   March 12, 2026

2026   ·   PINN   LLM   Multi-Agent   Physics-Informed   ICLR     ·   research  

ExpertFlow: 基于预测性专家缓存与令牌调度的高效MoE推理 | Efficient MoE Inference via Predictive Expert Caching and Token Scheduling

2 min read   ·   March 12, 2026

2026   ·   MoE   Inference   Expert-Caching   Token-Scheduling   DAC   LLM     ·   research  

【转载】ACM MM 投稿论文模板修改成投稿模式 - marsggbo

原文链接： https://zhuanlan.zhihu.com/p/491172953

1 min read   ·   November 13, 2025   ·   cnblogs.com

2025