Bin Wu
PhD Candidate — Computer Science
University College London (UCL AI Center, UCL NLP) · Supervised by Prof. Emine Yilmaz
Building efficient, adaptive, and robust LLM-based agentic systems through optimization, search, and evaluation.
I am a third-year PhD student at University College London, affiliated with the UCL Natural Language Processing Group (UCL NLP) and the UCL Centre for Artificial Intelligence, studying optimization, search, and evaluation of LLM-based agentic systems in complex and dynamic environments. My work aims to improve their efficiency, adaptability, and reliability, with experience spanning both academic research and industry collaboration at Bloomberg AI.
My research interests include multi-agentic AI systems, self-evolving agents, agent orchestration and collaboration, and agent search & evaluation. I am a Bloomberg Data Science PhD Fellow and have interned twice at Bloomberg AI Research in London.
Research on Robust Agentic Systems
My research focuses on building efficient, adaptive, and robust LLM-based agentic systems. I approach this from four interconnected directions.
Enhancing Agentic Systems in Complex & Dynamic Environments
How can we improve the efficiency and adaptability of LLM agents when tool environments change over time?
- Proposed a joint context optimization framework that reduces tool calls by up to 70% while maintaining effectiveness (ACL 2025).
- Proposed a continual documentation adaptation framework enabling LLM agents to self-evolve under dynamic tool environments without model retraining. (Under Review)
- Proposed a trajectory-based credit assignment framework for multi-agent prompt optimization. (Under Review)
AgentSearch: Indexing, Retrieval & Ranking of AI Agents
How can we systematically discover, represent, and retrieve the right AI agent for a given task — treating agents as first-class IR objects?
- Founded and led the SIGIR 2026 Workshop "AgentSearch@SIGIR 26: Indexing, Retrieval, and Ranking of AI Agents" (accepted).
- More works are coming soon.
Diagnosis & Evaluation of LLM and Agentic Systems
How do we rigorously diagnose failure modes and measure reliable progress in LLM and multi-agent systems?
- Investigated the mechanism of LLM personalization, revealing positional and compositional effects of user profiles (CIKM 2025).
- Investigated the retrieval-augmented question answering pipeline built on goal-oriented dialogues (Under Review).
Few-Shot Learning for Personalized Search & Recommendation
How can we transfer meta-knowledge across users and tasks to achieve strong performance under severe data sparsity?
- Developed a Bayesian online meta-learning framework for personalized product search under few-/zero-shot settings (WWW 2022).
- Built a dynamic Bayesian contrastive predictive coding model for structured user–product–query representation over time (TWeb 2023).
- Designed a compositional continual meta-learning algorithm with evidential sparsification for adaptive knowledge sharing across heterogeneous tasks (ICML 2023).
Publications
Preprints & Under Review
-
AgentSearch: Indexing, Retrieval, and Ranking of AI AgentsSIGIR 2026 Workshop Proposal arXiv
Selected Publications
-
A Joint Optimization Framework for Enhancing Efficiency of Tool Utilization in LLM Agents
@inproceedings{wu2025joint, title={A joint optimization framework for enhancing efficiency of tool utilization in llm agents}, author={Wu, Bin and Meij, Edgar and Yilmaz, Emine}, booktitle={Findings of the Association for Computational Linguistics: ACL 2025}, pages={22361--22373}, year={2025} } -
Empirical Analysis on User Profile in Personalized LLMsCIKM 2025 PDF
@inproceedings{wu2025empirical, title = {Empirical Analysis on User Profile in Personalized LLMs}, author = {Bin Wu and Zhengyan Shi and Hossein A Rahmani and Varsha Ramineni and Emine Yilmaz}, booktitle = {Proceedings of the 34th ACM International Conference on Information and Knowledge Management}, year = {2025} } -
Rethinking the Potential of Multimodality in Collaborative Problem Solving Diagnosis with Large Language ModelsAIED 2025 🏆 Best Student Paper PDF
@inproceedings{wong2025rethinking, title={Rethinking the Potential of Multimodality in Collaborative Problem Solving Diagnosis with Large Language Models}, author={Wong, Kester and Wu, Bin and Bulathwela, Sahan and Cukurova, Mutlu}, booktitle={International Conference on Artificial Intelligence in Education}, pages={18--32}, year={2025} } -
Instruction Tuning with Loss over InstructionsNeurIPS 2024 PDF
@article{shi2024instruction, title={Instruction tuning with loss over instructions}, author={Shi, Zhengyan and Yang, Adam X and Wu, Bin and Aitchison, Laurence and Yilmaz, Emine and Lipani, Aldo}, journal={Advances in Neural Information Processing Systems}, volume={37}, pages={69176--69205}, year={2024} } -
Adaptive Compositional Continual Meta-LearningICML 2023 PDF
@inproceedings{wu2023adaptive, title={Adaptive compositional continual meta-learning}, author={Wu, Bin and Fang, Jinyuan and Zeng, Xiangxiang and Liang, Shangsong and Zhang, Qiang}, booktitle={International Conference on Machine Learning}, pages={37358--37378}, year={2023}, organization={PMLR} } -
Dynamic Bayesian Contrastive Predictive Coding Model for Personalized Product SearchACM TWeb 2023 PDF
@article{wu2023dynamic, title={Dynamic Bayesian contrastive predictive coding model for personalized product search}, author={Wu, Bin and Meng, Zaiqiao and Liang, Shangsong}, journal={ACM Transactions on the Web}, volume={17}, number={4}, pages={1--31}, year={2023}, publisher={ACM New York, NY} } -
Meta-Learning Helps Personalized Product SearchWWW 2022 PDF
@inproceedings{wu2022meta, title={Meta-learning helps personalized product search}, author={Wu, Bin and Meng, Zaiqiao and Zhang, Qiang and Liang, Shangsong}, booktitle={Proceedings of the ACM Web Conference 2022}, pages={2277--2287}, year={2022} }
Education & Experience
Education
- Supervised by Prof. Emine Yilmaz & Dr. Edgar Meij
- Bloomberg Data Science PhD Fellow (2023–2025)
- Focus: optimization, search, and evaluation of LLM-based agentic systems
- Supervised by Prof. Shangsong Liang & Dr. Qiang Zhang
- Outstanding Graduate Student Award, 2022
- Focus: meta-learning, personalized search, and continual learning
Work Experience
- Building retrieval-augmented QA systems on goal-oriented dialogues.
- Leader & Mentor: Dr. Mohamed Yahya, Dr. Sawan Kumar, Dr. Prasetya Ajie Utama
- Building multi-dimensional automated evaluation system for contextual QA generation.
- Leader & Mentor: Dr. Mohamed Yahya, Dr. Sawan Kumar
News
- 🎉 SIGIR 2026 Workshop "AgentSearch: Indexing, Retrieval, and Ranking of AI Agents" accepted. Founding organizer.
- 🎉 New workshop at AAAI 2026: co-organizing "New Frontiers in Information Retrieval".
- 🏅 Awarded the NVIDIA Academic Grant Program — dual grants supporting LLM training (GPU-hours) and DGX machine for inference experiments.
- 🎉 Paper accepted at CIKM 2025.
- 📰 Tech At Bloomberg features our ACL 2025 paper on improved agent tool-calling methodology.
- 💼 Started 2nd research internship at Bloomberg AI, London
- 🎉 One Paper accepted at AIED 2025; wins 🏆 Best Student Paper Award.
- 🎉 Two Papers accepted at ACL 2025.
About
I am a third-year PhD candidate in Computer Science at University College London (UCL NLP Group), supervised by Prof. Emine Yilmaz and Dr. Edgar Meij. My research studies how to build efficient, adaptive, and robust LLM-based agentic systems that can handle complex and dynamic real-world environments.
Before my PhD, I completed an M.S. in Software Engineering at Sun Yat-sen University (supervised by Prof. Shangsong Liang and Dr. Qiang Zhang) and a B.S. in Software Engineering also at Sun Yat-sen University. I have interned twice at Bloomberg AI Research in London, working on QA system and automated evaluation systems that serve professional clients globally.
I am a Bloomberg Data Science PhD Fellow (2023–) and my research is additionally supported by NVIDIA and OpenAI. I believe in open, reproducible science and am always happy to discuss research. Feel free to reach out by email.
Download CV (PDF)🏆 Awards & Funding
- Bloomberg Data Science PhD Fellowship — Bloomberg, 2023, 2024, 2025
- NVIDIA Academic Grant Program — Nvidia, 2025 (GPU-hour allocation + DGX inference machine)
- Best Student Paper Award — AIED 2025
- OpenAI Researcher Access Program — OpenAI, 2024 ($3,000 API quota)
- Outstanding Graduate Student — Sun Yat-sen University, 2022
🎓 Academic Service
- Reviewer: WWW 2025–2026, ARR 2025–2026, NeurIPS 2025, ICLR 2026, ICML 2026, TOIS
- Workshop Organizer: AgentSearch @ SIGIR 2026, New Frontiers in Information Retrieval @ AAAI 2026
- Teaching Assistant: Information Retrieval & Data Mining (COMP0084), UCL (Spring 2024, 2025), Machine Learning for Data Science (CEGE0004), UCL (Spring 2024, 2025), Machine Learning and Data Mining, SYSU (Spring 2022)
