Yulin Hu
Hello! I am a master’s student in Computer Science and Technology at Harbin Institute of Technology, in the Sentiment Computing (SC) group of SCIR (Research Center for Social Computing and Interactive Robotics). I have been admitted to HIT’s Ph.D. program and will begin my doctoral studies in Spring 2027. I am advised by Prof. Yanyan Zhao and co-advised by Prof. Weixiang Zhao. I am currently a research intern at Huawei Consumer Business Group.
My research focuses on long-term memory for multimodal agents: how agents store, maintain, retrieve, and use information across sustained interactions. I also work on personalized and empathetic dialogue, safety alignment, and reasoning in large language models.
I am interested in building agents that understand users over time and use memory when it is relevant and helpful. Feel free to reach out at ylhu@ir.hit.edu.cn.
🔥 News
-
Nov. 2026 · Admitted to the Ph.D. program at Harbin Institute of Technology through the master’s-to-Ph.D. transfer pathway. I will begin my doctoral studies in Spring 2027.
-
Oct. 2026 · SC-Embedding is under review at WWW 2027, and How Do Language Models Use Memory? is under review at ICLR 2027. Code is available for both projects.
-
Sept. 2026 · CUE-Mem, our benchmark for long-term user memory from implicit multimodal cues, is now available on arXiv. Under review at AAAI 2027.
- 2026 · OP-Bench, our work on over-personalization in memory-augmented conversational agents, is accepted to EMNLP 2026.
- 2026 · Our work on proactive memory retrieval, tool-enhanced emotional support, and personalized-agent safety appears at ACL 2026 / Findings.
- 2025 · I started my master’s studies at Harbin Institute of Technology.
📝 Publications
🧠 Agent Memory & Personalization
- WWW 2027 (Under Review) SC-Embedding: Selective Contextualization for Long-Context Retrieval, First Author Memory Retrieval [Code] [Models]
- ICLR 2027 (Under Review) How Do Language Models Use Memory? From Internal Strategy Readout to Adaptive Control, First Author Memory Steering [Code]
- AAAI 2027 (Under Review) CUE-Mem: Benchmarking Long-Term User Memory via Implicit Cues in Multimodal Conversations, First Author Benchmark Memory Multimodal [Code] [Demo] [Dataset]
- EMNLP 2026 OP-Bench: Benchmarking Over-Personalization for Memory-Augmented Personalized Conversational Agents, First Author Benchmark Memory Personalization [Code] Citations: 29
- ACL 2026 (Findings) ENPMR-Bench: Benchmarking Proactive Memory Retrieval for Emotional Support Agents, Benchmark Memory Emotion Citations: 4
- arXiv 2026 SocialPersona: Benchmarking Personalized Profiling and Response with Multimodal Social-Media Context, Benchmark Multimodal Personalization
- NeurIPS 2025 Teaching Language Models to Evolve with Users: Dynamic Profile Modeling for Personalized Alignment, Co-first Author Personalization Alignment Citations: 38
🛡️ Safety Alignment & Role-playing
- ACL 2026 When Personalization Legitimizes Risks: Uncovering Safety Vulnerabilities in Personalized Dialogue Agents, Benchmark Safety Memory Personalization Citations: 5
- IJCAI 2026 Towards Comprehensive and Efficient Post Safety Alignment of Large Language Models via Safety Patching, Student First Author Safety Alignment [arXiv] Citations: 39
- Frontiers of Computer Science 2026 The Gains Do Not Make Up for the Losses: A Comprehensive Evaluation for Safety Alignment of Large Language Models via Machine Unlearning, Student First Author Benchmark Safety Unlearning Citations: 7
- ACL 2025 Beware of Your Po! Measuring and Mitigating AI Safety Risks in Role-Play Fine-Tuning of LLMs, Co-first Author Safety [Code] Citations: 36
- EMNLP 2025 (Oral) AdaSteer: Your Aligned LLM Is Inherently an Adaptive Jailbreak Defender, Safety Steering [Code] Citations: 58
- ACL 2025 (Oral & Panel · Top 0.3% of submissions) MPO: Multilingual Safety Alignment via Reward Gap Optimization, Student First Author Multilingual Safety [Code] Citations: 28
💬 Emotional Intelligence & Dialogue
- ACL 2026 TEA-Bench: A Systematic Benchmarking of Tool-enhanced Emotional Support Dialogue Agent, Benchmark Agent Emotion Tool Use Citations: 3
- EMNLP 2025 (Findings) Chain of Strategy Optimization Makes Large Language Models Better Emotional Supporter, Emotion Dialogue [Code] Citations: 27
- ACL 2024 (Findings) Both Matter: Enhancing the Emotional Intelligence of Large Language Models without Compromising the General Intelligence, Emotion Alignment Citations: 30
🎯 Reasoning & Efficiency
- AAAI 2026 Trade-offs in Large Reasoning Models: An Empirical Analysis of Deliberative and Adaptive Reasoning over Foundational Capabilities, Co-first Author Reasoning Evaluation [Code] Citations: 27
- arXiv 2025 Exploring and Exploiting the Inherent Efficiency within Large Reasoning Models for Self-Guided Efficiency Enhancement, Reasoning Efficiency Citations: 15
🌏 Multilingual Modelling & Reasoning
- NeurIPS 2025 (Spotlight) When Less Language Is More: Language-Reasoning Disentanglement Makes LLMs Better Multilingual Reasoners, Multilingual Reasoning Citations: 20
- arXiv 2024 LENS: Rethinking Multilingual Enhancement for Large Language Models, Student First Author Multilingual [Code] Citations: 11
📚 Continual Learning
- ACL 2024 SAPT: A Shared Attention Framework for Parameter-Efficient Continual Learning of Large Language Models, Continual Learning Efficiency [Code] 🔥 Citations: 119
🛠️ Research & Applications
- QiaoBan — Core developer of a personalized, empathetic conversational model for children and adolescents. Official news: QiaoBan-P1 release (June 2025).
- HIT AI Counselor — Core project lead and developer of a campus assistant combining university knowledge retrieval and empathetic dialogue. Project news: official launch (August 2025).
🎓 Education
- Spring 2027 (incoming) · Ph.D. studies, Harbin Institute of Technology. Admitted through the master’s-to-Ph.D. transfer pathway.
- Sept. 2025 – Present · Master’s student in Computer Science and Technology, Harbin Institute of Technology. Advisor: Prof. Yanyan Zhao; Co-advisor: Prof. Weixiang Zhao.
- Sept. 2021 – June 2025 · Bachelor’s degree in Computer Science and Technology, School of Future Technology, Harbin Institute of Technology.
💻 Research Internships
- Huawei Consumer Business Group · Research Intern · June 2026 – Present.
- Du Xiaoman Financial Technology · Research Intern · Dec. 2024 – Mar. 2025. Research on empathetic and role-based conversational language models.
🎖️ Honors & Awards
- National Scholarship for Graduate Students, Harbin Institute of Technology, 2026.
- University-level Outstanding Student Role Model, Harbin Institute of Technology, 2026.
- Special-class Graduate Academic Scholarship, Harbin Institute of Technology, 2025 & 2026.
- Outstanding Undergraduate Thesis, Harbin Institute of Technology, 2025.
- Wentian Scholarship (college-level), Harbin Institute of Technology, four consecutive years.
- Innovation and Entrepreneurship Scholarship (college-level), Harbin Institute of Technology, four consecutive years.
🤝 Academic Service
- Journal Reviewer: Science China Information Sciences (SCIS).
- Conference Reviewer: ICLR 2027, WWW 2027.