Sihao Liu | AI4GC Lab

About Me

Hi! I'm Sihao Liu (刘思浩), a master's student at AI4GC Lab, Zhejiang University, advised by Prof. Shengyu Zhang.

My research focuses on efficient inference for large language models, including KV cache compression and long-context optimization. I'm also interested in causal-aware sequential recommendation and scaling laws for recommendation models. Previously, I worked on semantic-disentangled 3D human reconstruction using Gaussian Splatting at Zhejiang Lab.

Feel free to reach out if you're interested in collaboration!

Research Interests

LLM Inference Optimization — KV cache compression, long-context reasoning acceleration, and efficient decoding for multimodal LLMs.
Causal Recommendation — Counterfactual intervention learning and causal dependency modeling in sequential recommendation.
Scaling Laws — Resource-optimized scaling laws for large-scale recommendation models with RetNet-based architectures.

Education

2024 – present · M.S. in Software Engineering, Zhejiang University
2020 – 2024 · B.S. in Software Engineering, Zhejiang University of Technology (GPA: 3.83/4.0, Rank: 1/98)

Selected Awards

Government Scholarship (3 consecutive years, 2020–2023)
Silver Award, National College Student Algorithm Design Competition (2024)
National Undergraduate Innovation and Entrepreneurship Training Program Award
Certified Software Designer & Systems Integration Engineer (National Certification)

Publications

ACL Findings2026

RetentiveKV: State-Space Memory for Uncertainty-Aware Multimodal KV Cache Eviction

Sihao Liu, YuFan Xiong, Zhonghua Jiang, Zhaode Wang, Shengyu Zhang^✉

Paper Blog

arXiv2025

PureKV: Plug-and-Play KV Cache Optimization with Spatial-Temporal Sparse Attention for Vision-Language Large Models

Zhonghua Jiang, Kunxi Li, Yiyun Zhou, Sihao Liu, Yufan Xiong, Zhaode Wang, Chengfei Lv, Shengyu Zhang^✉

Paper

arXiv2024

Gaussian Control with Hierarchical Semantic Graphs in 3D Human Recovery

Hongsheng Wang, Weiyue Zhang, Sihao Liu, Xinrui Zhou, Jing Li, Zhanyun Tang, Shengyu Zhang, Fei Wu, Feng Lin

JCRD2024

Lookahead Analysis and Discussion of Research Hotspots in Artificial Intelligence from 2021 to 2023

Zishu Wei, Yue Han, Sihao Liu, Shengyu Zhang, Fei Wu

Paper

JCST2024

Intent-aware Graph-level Embedding Learning Based Recommendation

Pengyi Hao, Sihao Liu, Cong Bai

Blog

On-site

Jun 2026

RetentiveKV: State-Space Memory for Uncertainty-Aware Multimodal KV Cache Eviction

RetentiveKV reformulates KV cache eviction from discrete context truncation to continuous memory evolution using entropy-driven State Space Models, achieving 5× KV cache compression and 1.5× decoding acceleration on multimodal benchmarks. Accepted to ACL 2026 Findings.