Shenglai Zeng

Education

Michigan State University

Ph.D. in Computer Science & Engineering · 2023 ~ Present

DSE Lab · Advisor: Prof. Jiliang Tang

University of Electronic Science and Technology of China

B.Sc. in Computer Science & Engineering · 2019 ~ 2023

Yingcai Honor School · CGPA 3.98/4.00 (Rank 1st among 100 students)

Experience

Meta AI

Summer 2026 · Seattle, WA

Research Scientist Intern · interaction model & multi-modal proactive agent
Mentors: Yang Xiao, Deren Lei, Kai Sun, Nanshu Wang, Xilun Chen, Xinyuan Zhang, Scott Yih, Xin Luna Dong
Working on interaction model & next-generation multi-modal proactive agent

Amazon (Search Science / Rufus)

Summer 2024 & 2025 · Palo Alto, CA

Applied Scientist Intern · Retrieval-Augmented Generation & Personalization
Mentors: Tianqi Zheng, Dante Everaert · Managers: Hanqing Lu, Chuan Tian, Monica Xiao Cheng
Outcomes: Knowledge Checking in RAG (NAACL 2025 Oral), Robust RAG (ACL 2025), Synthetic & Multi-modal RAG (EMNLP 2025), Efficient Personalized LLMs (ACL 2026 Oral)

Baidu Inc. (Search Science)

May 2023 ~ May 2024 · Beijing, China

Research Intern · Memorization & Privacy of LLMs
Mentors: Yiding Liu, Dawei Yin
Outcomes: LLM Memorization (ACL 2024), RAG Privacy (ACL 2024)

University of British Columbia

Summer 2022 · Vancouver, Canada

Research Intern (MITACS) · Federated Data Evaluation with Unlearning
Mentor: Prof. Xiaoxiao Li

The Pennsylvania State University

2021 ~ 2022 · Remote

Research Intern · Semi-supervised Federated Learning
Mentor: Prof. Fenglong Ma · Outcome: SDM 2023

The University of Chicago

Mar 2020 ~ Mar 2021 · Remote

Online Intern · IoT & Sensing Security
Mentor: Shinan Liu

News

Publications

* equal contribution. Full list on Google Scholar (1000+ citations).

• Preprints
• Magnifying What Matters: Attention-Guided Adaptive Rendering for Visual Text Comprehension
  Submitted to NeurIPS 2026
  Shenglai Zeng, Qirui Wang, Kai Guo, Xinnan Dai, Xianxuan Long, Hui Liu
  Key Insight: Attention-guided adaptive rendering sharpens visual text comprehension in multi-modal LLMs.
  MultimodalLLMEfficiencyData/Context
• Beyond Static Retrieval: Opportunities and Pitfalls of Iterative Retrieval in GraphRAG
  Submitted to ACL-ARR
  Kai Guo, Xinnan Dai, Shenglai Zeng, Harry Shomer, Haoyu Han, Yu Wang, Jiliang Tang
  Key Insight: Characterizes when iterative retrieval helps versus hurts in GraphRAG.
  Paper
  RAGLLMIRFactualityGraph
• GraphGhosts: Tracing Reasoning Structures Behind Large Language Models
  Submitted to ACL-ARR
  Xinnan Dai, Kai Guo, Chung-Hsiang Lo, Shenglai Zeng, Jiayuan Ding, Dongsheng Luo, Subhabrata Mukherjee, Jiliang Tang
  Key Insight: Traces the latent reasoning structures of LLMs through a graph lens.
  LLMReasoningInterpretabilityGraph
• Structural Alignment Faking: Hiding Malicious Capabilities in Mixture-of-Experts LLMs
  Preprint
  Yuping Lin, Pengfei He, Haoran Zhao, Shenglai Zeng, Zhen Xiang, Hui Liu, Jiliang Tang
  Key Insight: Reveals how Mixture-of-Experts LLMs can hide malicious capabilities via structural alignment faking.
  SafetyLLMPost-training
• MemTrace: Probing What Final Accuracy Misses in Long-Term Memory
  Preprint
  Xianxuan Long, Zhikai Chen, Shenglai Zeng, Shouren Wang, Kai Guo, Jiliang Tang
  Key Insight: Probes long-term memory failures that final-accuracy metrics overlook.
  Paper
  AgentMemoryLLMIRBenchmarkAgent Harness
• Exploring Cross-Scenario Generality of Agentic Memory Systems: Diagnostics and a Strong Baseline
  Preprint
  Zhikai Chen, Jialiang Gu, Junyu Yin, Xianxuan Long, Shenglai Zeng, Xiaoze Liu, Kai Guo, Keren Zhou, Jiliang Tang
  Key Insight: Diagnoses the cross-scenario generality of agentic memory systems and proposes a strong baseline.
  Paper
  AgentMemoryLLMIRAgent HarnessBenchmark
• Uncovering Graph Reasoning in Decoder-only Transformers with Circuit Tracing
  Preprint
  Xinnan Dai, Chung-Hsiang Lo, Kai Guo, Shenglai Zeng, Dongsheng Luo, Jiliang Tang
  Key Insight: Circuit tracing reveals how decoder-only transformers perform graph reasoning.
  Paper
  LLMReasoningInterpretabilityGraph
• Copyright Protection in Generative AI: A Technical Perspective
  Preprint
  Jie Ren, Han Xu, Pengfei He, Yingqian Cui, Shenglai Zeng, Jiankun Zhang, Hongzhi Wen, Jiayuan Ding, Hui Liu, Yi Chang, Jiliang Tang
  Key Insight: A technical survey of copyright protection methods for generative AI.
  Paper
  SafetySurveyLLM
• Conference & Journal
• Fix Before Search: Benchmarking Agentic Query Visual Pre-processing in Multimodal Retrieval-augmented Generation
  ICML 2026
  Shenglai Zeng*, Jiankun Zhang*, Kai Guo, Xinnan Dai, Hui Liu, Jiliang Tang, Yi Chang
  Key Insight: Agentic visual pre-processing of queries before retrieval boosts multimodal RAG.
  Paper
  RAGAgentMultimodalLLMIRAgent HarnessPost-trainingBenchmarkFactuality
• Attn-GS: Attention-Guided Context Compression for Efficient Personalized LLMs
  ACL 2026 Main Oral Best Paper Candidate
  Shenglai Zeng, Tianqi Zheng, Chuan Tian, Dante Everaert, Yau-Shian Wang, Yupin Huang, Michael J. Morais, Rohit Patki, Jinjin Tian, Xinnan Dai, Kai Guo, Monica Xiao Cheng, Hui Liu
  Key Insight: Attention-guided context compression makes personalized LLMs efficient without losing quality.
  Paper
  LLMEfficiencyPersonalizationInterpretabilityData/Context
• When Do Hallucinations Arise? A Graph Perspective on the Evolution of Path Reuse and Path Compression
  ICML 2026
  Xinnan Dai, Kai Yang, Cheng Luo, Shenglai Zeng, Kai Guo, Jiliang Tang
  Key Insight: A graph view of how path reuse and compression give rise to hallucination.
  Paper
  LLMReasoningInterpretabilityFactualityGraph
• Why Retrieval-Augmented Generation Fails: A Graph Perspective
  KDD 2026
  Kai Guo, Xinnan Dai, Zhibo Zhang, Nuohan Lin, Shenglai Zeng, Jie Ren, Haoyu Han, Jiliang Tang
  Key Insight: Diagnoses why RAG fails through a graph-structured analysis.
  Paper
  RAGLLMIRInterpretabilityFactualityGraph
• Keeping an Eye on LLM Unlearning: The Hidden Risk and Remedy
  NeurIPS 2025
  Jie Ren, Zhenwei Dai, Xianfeng Tang, Yue Xing, Shenglai Zeng, Hui Liu, Jingying Zeng, Qiankun Peng, Samarth Varshney, Suhang Wang, Qi He, Charu C. Aggarwal, Hui Liu
  Key Insight: Exposes hidden risks in LLM unlearning and proposes a remedy.
  Paper
  SafetyLLMPost-trainingBenchmark
• Mitigating the Privacy Issues in Retrieval-Augmented Generation (RAG) via Pure Synthetic Data
  EMNLP 2025 Main
  Shenglai Zeng, Jiankun Zhang, Pengfei He, Jie Ren, Tianqi Zheng, Hanqing Lu, Han Xu, Hui Liu, Yue Xing, Jiliang Tang
  Key Insight: Pure synthetic data mitigates retrieval-data privacy leakage in RAG.
  Paper
  RAGPrivacyLLMIRData/Context
• Beyond Text: Unveiling Privacy Vulnerabilities in Multi-modal Retrieval-Augmented Generation
  EMNLP 2025 Main
  Jiankun Zhang*, Shenglai Zeng*, Jie Ren, Tianqi Zheng, Hui Liu, Xianfeng Tang, Yi Chang
  Key Insight: Uncovers privacy vulnerabilities unique to multi-modal RAG.
  Paper
  RAGPrivacyMultimodalLLMIR
• Empowering GraphRAG with Knowledge Filtering and Integration
  EMNLP 2025 Main
  Kai Guo, Harry Shomer, Shenglai Zeng, Haoyu Han, Yu Wang, Jiliang Tang
  Key Insight: Knowledge filtering and integration substantially empower GraphRAG.
  Paper
  RAGLLMIRFactualityGraph
• Towards Context-Robust LLMs: A Gated Representation Fine-tuning Approach
  ACL 2025 Main
  Shenglai Zeng, Pengfei He, Kai Guo, Tianqi Zheng, Hanqing Lu, Yue Xing, Hui Liu
  Key Insight: Gated representation fine-tuning makes LLMs robust to noisy context.
  Paper
  RAGLLMIRPost-trainingFactuality
• Unveiling Privacy Risks in LLM Agent Memory
  ACL 2025 Main
  Bo Wang, Weiyi He, Pengfei He, Shenglai Zeng, Zhen Xiang, Yue Xing, Jiliang Tang
  Key Insight: Reveals privacy risks hidden in the memory of LLM agents.
  Paper
  PrivacyAgentMemoryLLMIRAgent Harness
• Unveiling Mode Connectivity in Graph Neural Networks
  KDD 2025
  Bingheng Li, Zhikai Chen, Haoyu Han, Shenglai Zeng, Jingzhe Liu, Jiliang Tang
  Key Insight: Studies mode connectivity in the loss landscape of graph neural networks.
  Paper
  Graph
• Towards Knowledge Checking in Retrieval-Augmented Generation: A Representation Perspective
  NAACL 2025 Oral
  Shenglai Zeng, Jiankun Zhang, Bingheng Li, Yuping Lin, Tianqi Zheng, Dante Everaert, Hanqing Lu, Hui Liu, Yue Xing, Monica Xiao Cheng, Jiliang Tang
  Key Insight: Uses LLM internal representations to check knowledge sufficiency in RAG.
  Paper
  RAGLLMIRInterpretabilityData/ContextFactuality
• Six-CD: Benchmarking Concept Removals for Benign Text-to-Image Diffusion Models
  CVPR 2025
  Jie Ren, Kangrui Chen, Yingqian Cui, Shenglai Zeng, Hui Liu, Yue Xing, Jiliang Tang, Lingjuan Lyu
  Key Insight: A comprehensive benchmark for concept removal in text-to-image diffusion models.
  Paper
  DiffusionSafetyBenchmark
• Stealthy Backdoor Attack via Confidence-Driven Sampling
  TMLR
  Pengfei He, Yue Xing, Han Xu, Jie Ren, Yingqian Cui, Shenglai Zeng, Jiliang Tang, Makoto Yamada, Mohammad Sabokrou
  Key Insight: Confidence-driven sampling enables stealthier, more resilient backdoor attacks.
  Paper
  Safety
• On the Generalization of Training-based ChatGPT Detection Methods
  EMNLP 2024 Findings
  Han Xu, Jie Ren, Pengfei He, Yingqian Cui, Shenglai Zeng, Hui Liu, Jiliang Tang, Amy Liu
  Key Insight: Examines the generalization limits of training-based ChatGPT detectors.
  Paper
  LLMSafety
• Unveiling and Mitigating Memorization in Text-to-Image Diffusion Models through Cross Attention
  ECCV 2024
  Jie Ren, Yaxin Li, Shenglai Zeng, Han Xu, Lingjuan Lyu, Yue Xing, Jiliang Tang
  Key Insight: Cross-attention drives memorization in diffusion models, and can be edited to mitigate it.
  Paper
  DiffusionMemoryPrivacy
• Exploring Memorization in Fine-tuned Language Models
  ACL 2024 Main
  Shenglai Zeng*, Yaxin Li*, Jie Ren, Yiding Liu, Han Xu, Pengfei He, Yue Xing, Shuaiqiang Wang, Jiliang Tang, Dawei Yin
  Key Insight: Characterizes how and when fine-tuned language models memorize training data.
  Paper
  MemoryPrivacyLLMPost-training
• The Good and The Bad: Exploring Privacy Issues in Retrieval-Augmented Generation (RAG)
  ACL 2024 Findings
  Shenglai Zeng*, Jiankun Zhang*, Pengfei He, Yue Xing, Yiding Liu, Han Xu, Jie Ren, Shuaiqiang Wang, Dawei Yin, Yi Chang, Jiliang Tang
  Key Insight: First study exposing retrieval-data leakage as a core privacy risk of RAG.
  Paper
  RAGPrivacyLLMIR
• Sharpness-Aware Data Poisoning Attack
  ICLR 2024 Spotlight
  Pengfei He, Han Xu, Jie Ren, Yingqian Cui, Shenglai Zeng, Hui Liu, Charu C. Aggarwal, Jiliang Tang
  Key Insight: Sharpness-aware data poisoning that remains effective against common defenses.
  Paper
  Safety
• Evaluating Graph Neural Networks for Link Prediction: Current Pitfalls and New Benchmarking
  NeurIPS 2023 (Benchmark)
  Juanhui Li, Harry Shomer, Haitao Mao, Shenglai Zeng, Yao Ma, Neil Shah, Jiliang Tang, Dawei Yin
  Key Insight: Exposes evaluation pitfalls and offers new benchmarks for GNN link prediction.
  Paper
  BenchmarkGraph
• HFedMS: Heterogeneous Federated Learning with Memorable Data Semantics in Industrial Metaverse
  IEEE Transactions on Cloud Computing 2023 Best Paper Award
  Shenglai Zeng, Zonghang Li, Hongfang Yu, Zhihao Zhang, Long Luo, Bo Li, Dusit Niyato
  Key Insight: Heterogeneous federated learning with memorable data semantics for the industrial metaverse.
  Paper
  Distributed MLPersonalizationEfficiency
• Heterogeneous Federated Learning via Grouped Sequential-to-Parallel Training
  DASFAA 2022
  Shenglai Zeng, Zonghang Li, Hongfang Yu, Yihong He, Zenglin Xu, Dusit Niyato, Han Yu
  Key Insight: Grouped sequential-to-parallel training for heterogeneous federated learning.
  Paper Video
  Distributed MLPersonalizationEfficiency
• Knowledge-Enhanced Semi-Supervised Federated Learning for Aggregating Heterogeneous Lightweight Clients in IoT
  SDM 2023
  Jiaqi Wang*, Shenglai Zeng*, Zewei Long, Yaqing Wang, Houping Xiao, Fenglong Ma
  Key Insight: Knowledge-enhanced semi-supervised FL aggregates heterogeneous lightweight IoT clients.
  Paper
  Distributed MLPersonalizationEfficiency

Awards & Service