Hi! I’m Jia-Huei (Dylan) Ju, a PhD student at Johns Hopkins University, co-advised by Dr. Andrew Yates and Prof. Benjamin Van Durme. I started my PhD in 2024 at the IRLab at University of Amsterdam, where I was supported by Hybrid Intelligence Centre and advisored by Dr. Andrew Yates, Prof. Suzan Verberne and Prof. Maarten de Rijke, before transferring to JHU.
My research is in information retrieval and natural language processing, with a focus on retrieval-augmented generation (RAG). I design retrieval systems that efficiently augment knowledge for agents. I’m particularly interested in how knowledge should be represented, and how human can use and acquire knowledge more efficiently.
Currently, I study how retrieval should manage knowledge for LLMs acting as lifelong agents. I focus on three properties of retrieved context: information coverage, so that the context as a whole covers all the knowledge a task needs; context editability, so that it stays up to date as information and information needs change; and knowledge reusability, so that it can serve as a reusable memory shared across tasks, sessions, and agents.
In addition to these directions, I’m also interested in exploring cognitive science, neuroscience, and other interdisciplinary problems where retrieval can be part of the solution. If you’re working on a problem where retrieval might help. I’d be happy to brainstorm and chat! Feel free to reach out at jju8[at]jh.edu.
Publications
- SemBridge: Language Transfer in Sparse Encoders via Multilingual Semantic Bridge
- Search for Coverage: Learning Coverage-Aware Retrieval with Augmented Sub-Question Answerability
- LANCER: LLM Reranking for Nugget Coverage
- The Powerless Noise: How Experimental Settings Shape the Reported Power of Noise
- MILCO: Learned Sparse Retrieval Across Languages via a Multilingual Connector
- Better than Dense? Investigating the Natural Backward Compatibility of Learned Sparse Representations
- SERVAL: Surprisingly Effective Zero-Shot Visual Document Retrieval Powered by Large Vision and Language Models
- Controlled Retrieval-Augmented Context Evaluation for Long-form RAG
- Reproducibility, Replicability, and Insights into Visual Document Retrieval with Late Interaction
- Relevance-aware Diverse Query Generation for Out-of-domain Text Ranking
- IRLab-AMS at TREC'24 NeuCLIR Track
- Improving Conversational Passage Re-ranking with View Ensemble
- A Compare-and-contrast Multistage Pipeline for Uncovering Financial Signals in Financial Reports
- CFDA & CLIP Labs at TREC'23 Product Search Track
- CFDA & CLIP at TREC 2022 Conversational Assistance Track (CAsT)
- CFDA & CLIP at TREC 2022 NeuCLIR Track
- Text-to-text Multi-view Learning for Passage Re-ranking
- Paraphrase Query Expansion and Multi-view Point-wise Ranking
Also see my Google Scholar profile.