An Zhang | Professor, USTC

Biography

I am a Professor at the University of Science and Technology of China (USTC), where I lead the Alpha Lab. Prior to that, I was a Postdoctoral Research Fellow in NExT++ Lab, National University of Singapore, working with Prof. Tat-Seng Chua. I obtained my Ph.D. from the National University of Singapore in 2021 and joined USTC in 2025.

At Alpha Lab, my colleagues, students, and collaborators work together to develop foundation models (large language models & diffusion models) and AI Agents, with an emphasis on understanding the capabilities and properties of next-generation general AI models. Our research is motivated by, and contributes to, real-world applications in personalization (e.g., recommender systems), education (e.g., smart campuses), and safety (e.g., LLM Safety).

Research Interests

Personalization

LLM & diffusion-powered recommender systems and next-gen personalized AI.

LLM as Rec Diffusion Rec GenRec

Agentic LLM

Building autonomous AI agents with planning, tool-use, memory, and long-horizon decision-making.

AI Agents Tool Use Memory

LLM Reasoning

Advancing chain-of-thought, self-verification, and deep reasoning capabilities in LLMs.

Chain-of-Thought Self-Verify Long Reasoning

Trustworthy LLM

LLM safety, alignment, and robust AI systems for safe real-world deployment.

Safety Alignment Robustness

Personalization

Tutorials: [WWW 2025 Generative Recommendation Models] [CIKM 2025]
LLM as Recommender: [ReRe] [NeurIPS 2024 S-DPO] [SIGIR 2024 Agent4Rec]
LLM-enhanced Recommender: [SIGIR 2025 AlphaFuse] [KDD 2025 LLM2Rec] [ICLR 2025 AlphaRec]
Diffusion as Recommender: [NeurIPS 2025 PrefGrow] [ICLR 2025 PrefDiff] [iDreamRec]

Agentic LLM

Tutorials: [WWW 2024 Tutorial] [SIGIR 2024 Tutorial]
Agentic Systems: [NeurIPS 2025 AutoRefine] [NeurIPS 2025 AgentRecBench]
Memory & Long-Context: [GruMem] [LD-Agent] [MemOCR]
Multimodality: [ICLR 2025 FiSAO]

LLM Reasoning

Reasoning Strength: [NeurIPS 2025 AlphaReasoner] [NeurIPS 2025 Reasoning Strength Planning]
Self-Verification: [Learning-to-Self-Verify]
RAG & Knowledge: [NeurIPS 2025 Search-and-Refine]

Trustworthy LLM

Safety & Alignment: [AlphaAlign] [NeurIPS 2025 SSA] [NeurIPS 2025 RSafe] [AlphaSteer]
Agent Safety: [Risky-Bench] [NeurIPS 2024 ALI-Agent]
LLM Alignment: [NeurIPS 2024 S-DPO] [NeurIPS 2025 SSafe]

Research Projects

Open-source projects from Alpha Lab & collaborators — View Lab Page →

Agentic LLM & Reasoning

MemOCR

Visual Memory Agent

OCR-driven visual memory agent enabling structured long-term memory retrieval for AI systems.

⭐ -- GitHub

LD-Agent

Long-term Dialogue

Long-term dialogue agent with event-driven memory, persona extraction, and structured response generation.

⭐ -- GitHub

Learning-to-Self-Verify

LLM Reasoning

Teaching LLMs to self-verify their own outputs, improving both generation quality and verification accuracy.

⭐ -- GitHub

GruMem

Long-Context Memory

Gated recurrent memory for LLMs — up to 400% inference speedup on long-context reasoning tasks.

Project Page

Personalization

AlphaRec

Language-based Rec

Language representations as the foundation for collaborative filtering — lightweight, fast convergence, superior zero-shot abilities. ICLR 2025 Oral.

⭐ -- GitHub

MiniOneRec

Generative Recommendation

A minimalist one-stage generative recommendation framework powered by LLMs.

⭐ -- GitHub

Agent4Rec

Recommendation Agent

LLM-powered agents for simulating user behavior and evaluating recommender systems. SIGIR 2024.

⭐ -- GitHub

S-DPO

LLM Alignment

Softmax Direct Preference Optimization for aligning LLMs as recommenders. NeurIPS 2024.

⭐ -- GitHub

Trustworthy AI

AlphaSteer

LLM Safety

Learning refusal steering with principled null-space constraints — enhances safety without compromising utility. ICLR 2026.

⭐ -- GitHub

AlphaAlign

LLM Alignment

Incentivizing safety alignment through extremely simplified reinforcement learning with dual-component reward. ICLR 2026.

⭐ -- GitHub

Risky-Bench

Agent Safety

Benchmark for evaluating safety and risk-awareness of LLM-powered agents in real-world scenarios.

⭐ -- GitHub

News & Highlights

2026 / 01
🎉 Five papers accepted at ICLR 2026!
Big congrats to Wenyu Mao, Junkang Wu, Yaorui Shi, Yi Zhang, Leheng Sheng and all collaborators!

AlphaSteer: Learning Refusal Steering with Principled Null-Space Constraint AlphaAlign: Incentivizing Safety Alignment with Extremely Simplified Reinforcement Learning Quantile Advantage Estimation: Stabilizing RLVR for LLM Reasoning Look Back to Reason Forward: Revisitable Memory for Long-Context LLM Agents Denoising Neural Reranker for Recommender Systems
2025 / 09
🎉 Seven papers accepted at NeurIPS 2025!
Big congrats to Guoqing Hu, Yaorui Shi, Leheng Sheng, Chenhang Cui, Jingnan Zheng, Yuxin Chen, Yu Shang and all collaborators!

Fading to Grow: Growing Preference Ratios via Preference Fading Discrete Diffusion for Recommendation Search and Refine During Think: Facilitating Knowledge Refinement for Improved Retrieval-Augmented Reasoning On Reasoning Strength Planning in Large Reasoning Models Safe + Safe = Unsafe? Exploring How Safe Images Can Be Exploited to Jailbreak Large Vision-Language Models RSafe: Incentivizing Proactive Reasoning to Build Robust and Adaptive LLM Safeguards The Emergence of Abstract Thought in Large Language Models Beyond Any Language AgentRecBench: Benchmarking LLM Agent-based Personalized Recommender Systems
2025 / 06
One paper accepted at KDD 2025!
Big congrats to Yingzhi He and Xiaohao Liu!

LLM2Rec: Large Language Models Are Powerful Embedding Models for Sequential Recommendation
2025 / 04
One paper accepted at SIGIR 2025!
Big congrats to Guoqing Hu!

AlphaFuse: Learn ID Embeddings for Sequential Recommendation in Null Space of Language Embeddings
2025 / 01
Three papers accepted at ICLR 2025!
Congratulations to Leheng Sheng, Chenhang Cui, Shuo Liu and all collaborators!

Language Representations Can be What Recommenders Need: Findings and Potentials (⭐ Oral) Fine-Grained Verifiers: Preference Modeling as Next-token Prediction in Vision-Language Alignment Preference Diffusion for Recommendation

Join Alpha Lab

🎓 I am actively looking for highly motivated PhD students, master students, and interns to work together on LLM-powered agents, personalization, trustworthy AI, and their real-world applications. Please feel free to send me your CV and transcripts if you are interested. We are also seeking opportunities in research, partnership, and commercialization in exciting AI projects.

📧 Contact: An_Zhang@ustc.edu.cn

Professional Services

Program Committee member for NeurIPS and ACL; reviewer / PC for other major AI, ML, NLP and Data Mining venues including ICML, ICLR, KDD, WWW/TheWebConf, SIGIR, EMNLP, AAAI, WSDM, CVPR, ICCV, ECCV, ACM MM, as well as related workshops and journal tracks.

Honors & Awards

Rising Stars of Women In the Web Award, 2025
SIGIR Best PC Member Award, 2025