Jizhihui Liu

Academic Homepage

Jizhihui Liu

Undergraduate Student in Artificial Intelligence at Harbin Institute of Technology, Shenzhen

I am an undergraduate student in the Artificial Intelligence Academician Class, School of Future Technology, Harbin Institute of Technology. My current research focuses on Embodied AI, Multimodal Large Language Models, and efficient vision-language reasoning for robotic manipulation.

Email Me GitHub
2023112090@stu.hit.edu.cn Danielement321 Shenzhen, China HIT, School of Future Technology
Portrait of Jizhihui Liu

Profile

About

I am currently pursuing a B.Eng. in the Artificial Intelligence Academician Class at the School of Future Technology, Harbin Institute of Technology from September 2023 to June 2027.

My research lies at the intersection of Embodied AI and MLLMs, with ongoing work on vision-language models, inference acceleration, token compression, and robot manipulation.

Program B.Eng. in Artificial Intelligence
Institution Harbin Institute of Technology, Shenzhen
Research Areas Embodied AI, MLLMs, VLM Efficiency
Program Intro View Introduction

Research Output

Selected Publications

* Equal contribution, † Corresponding author

  1. FM2S: Towards Spatially-Correlated Noise Modeling in Zero-Shot Fluorescence Microscopy Image Denoising Accepted · MIR, JCR Q1

    Jizhihui Liu*, Qixun Teng*, Qing Ma, Junhui Hou, Junjun Jiang†

    Image Denoising · Low-Level Vision · Fluorescence Microscopy Imaging

    Springer arXiv
  2. HiPrune: Hierarchical Attention for Efficient Token Pruning in Vision-Language Models Accepted · AAAI-26 SAPP / ACL-26 Findings

    Jizhihui Liu*, Feiyi Du*, Guangdao Zhu*, Niu Lian, Jun Li, Bin Chen†, Weili Guan, Yaowei Wang

    Vision-Language Models · Token Pruning · Inference Acceleration

    arXiv
  3. TGV-KV: Text-Grounded KV Eviction for Vision-Language Models Submitted · ICML-26

    Jizhihui Liu, Ruizi Han, Miao Zhang†, Rui Shao, Xuebo Liu, Weili Guan, Yaowei Wang

    Vision-Language Models · KV Cache Eviction · Inference Acceleration

  4. ConsisVLA-4D: Advancing Spatiotemporal Consistency in Efficient 3D-Perception and 4D-Reasoning for Robotic Manipulation Accepted · CVPR-26

    Wei Li, Jizhihui Liu, Li Yixing, Junwen Tong, Rui Shao†, Liqiang Nie

    Embodied AI · VLA · VGGT · World Model

  5. H-GAR: A Hierarchical Interaction Framework via Goal-Driven Observation-Action Refinement for Robotic Manipulation Accepted · AAAI-26 Oral

    Yijie Zhu, Zitong YU, Ziyang Liu, Jie He, Jizhihui Liu, Jiuru Wang, Rui Shao†

    Embodied AI · VLA · Goal-Conditioned Action Generation

    arXiv

Projects

Research Experience

Embodied AI and VLA Large Model Research

Mar. 2025 - Feb. 2026

Research Intern | Advisor: Rui Shao

  • VLA Survey: Systematically organized representative VLA works. GitHub Repository with 300+ stars, later submitted to TPAMI.
  • ConsisVLA-4D: Introduced VGGT into robot action planning with future-state prediction for key regions, and took responsibility for the full implementation pipeline.
  • H-GAR: Explored a unified framework for future-state prediction and fine-grained action generation in robotic manipulation.

Token Compression and Practice for MLLMs

Dec. 2024 - Aug. 2025

Project Leader | Advisors: Bin Chen, Yaowei Wang

  • Analyzed the internal attention distribution of mainstream visual encoders such as CLIP and verified the correlation between attention and semantic objects.
  • Designed an efficient visual token pruning strategy that achieved 2.5x acceleration in the prefilling stage of large models.
  • Preliminary results were accepted by AAAI-26 Student Abstract and Poster Program, with the full paper later submitted to ACL-26.

Fluorescence Microscopy Image Denoising via Zero-Shot Learning

Sep. 2023 - Aug. 2024

Project Leader | Advisor: Junjun Jiang

  • Proposed an adaptive noise addition module tailored to Poisson-Gaussian noise for high-quality sample expansion.
  • Built a lightweight fully convolutional neural network that completed generation, training, and inference within 10 seconds on a single RTX 4070 Super GPU.
  • The project results were published in Machine Intelligence Research.

Activities

Academic Exchange

The 40th Annual AAAI Conference on Artificial Intelligence

Jan. 2026

Singapore

Oral and poster presentation for the Student Abstract and Poster Program.

Summer Course at National University of Singapore

Aug. 2025

Singapore

Completed the course Generative AI for Data Analytics: Unleashing Insights with ChatGPT.