Hello, I am Zeru, a first year Ph.D. student in Computer Science at Rutgers University. I am honored to be guided by Prof.Ruixiang Tang. I received my B.E. degree in Software Engineering at Dalian University of Technology

My research interests mainly focus on:

• LLM/VLM Reasoning and Post-training: I am highly interested in post-training for LLMs and VLMs. My focus is on developing more effective reinforcement learning algorithms, as well as leveraging simple and interpretable methods, to improve reasoning capabilities through post-training.
• Agentic AI: I am also interested in Agentic AI, especially in the design of agent memory. I am particularly drawn to a storage centric perspective, where memory is treated as a first-class component. My goal is to build principled memory systems that can significantly enhance agents’ capabilities in complex reasoning and real world problem solving.

My CV is here: Zeru’s CV

I am always open to research collaborations. If you are interested in my previous work or would like to discuss potential ideas, please feel free to contact me.

🔥 News

📷 Publication

ICML2026

A Single Layer to Explain Them All: Understanding Massive Values in Large Language Models

[ICML 2026]

Zeru Shi, Zhenting Wang, Fan Yang, Qifan Wang, Ruixiang Tang

TL;DR: We explored the actionable mechanistic interpretation of massive values in LLMs and propose a method to mitigate the massive activatons.

[Project Page]   [Arxiv]  [GitHub]

ICLR2025

From Commands to Prompts: LLM-based Semantic File System

[ICLR 2025]

TL;DR: We propose a vector-based agent memory system that enables users to manage and interact with computer files through natural language, eliminating the need for traditional Linux commands.

Zeru Shi*, Kai Mei*, Mingyu Jin, Yongye Su, Chaoji Zuo, Wenyue Hua, Wujiang Xu, Yujie Ren, Zirui Liu, Mengnan Du, Dong Deng, Yongfeng Zhang

[Arxiv] 
[GitHub]

IEEE TCSVT

SeFENet: Robust Deep Homography Estimation via Semantic-Driven Feature Enhancement

[IEEE TCSVT, IF=11.1]

TL;DR: We design a meta-learning framework to improve the robustness and performance of homography estimation under challenging environments.

Zeru Shi, Zengxi Zhang, Kemeng Cui, Ruizhe An, Jinyuan Liu, Zhiying Jiang

[Arxiv]

EMNLP2025 (Main)

Castle: Causal Cascade Updates in Relational Databases with Large Language Models

[EMNLP 2026, Main]

Yongye Su, Yucheng Zhang, Zeru Shi, Bruno Ribeiro, Elisa Bertino

[Arxiv]

📝 Pre-Print

Arxiv

Meaningless Tokens, Meaningful Gains: How Activation Shifts Enhance LLM Reasoning
Zeru Shi, Yingjia Wan, Zhenting Wang, Qifan Wang, Fan Yang, Elisa Kreiss, Ruixiang Tang

[Arxiv], [GitHub]

Arxiv

Auto-Prompt Generation is Not Robust:Prompt Optimization Driven by Pseudo Gradient
Zeru Shi, Zhenting Wang, Yongye Su, Weidi Luo, Hang Gao, Fan Yang, Ruixiang Tang, Yongfeng Zhang

[Arxiv], [GitHub]

🎖 Honors and Awards

• 2025.10 Rutgers Climate and Energy Institute (RCEI) Fellowship, USA
• 2025.6 Dalian excellent undergraduate graduates, China
• 2024.11 National Scholarship, China
• 2023.9 Special scholarship of NOK Corporation, Japan

📖 Educations

• 2021.09 - 2025.06, Dalian University of Technology, ISE, Liaoning, Undergraduate
• 2025.09 - Present, Rutgers University, Computer Science, New Brunswick, Ph.D.

💻 Internships

• 2026.06 - 2026.08, NVIDIA(Incoming), Santa Clara, CA, USA
• 2024.10 - 2025.04, Shanghai AI Laboratory, Shanghai, China
• 2024.02 - 2024.03, Li Auto.Inc, Beijing, China

🕵️ Services

• Conference Reviewer: ICLR 2026, ICML 2026, ARR, Neurips 2026