Orion Weller
I’m a fourth-year PhD student at the Center for Language and Speech Processing at Johns Hopkins University, advised by Benjamin Van Durme and Dawn Lawrie. I am broadly interested in natural language processing (NLP), information retrieval (IR), and machine learning (ML). My research is graciously supported by a NSF Graduate Research Fellowship.
My current research interests are situated between the NLP and IR fields, where I work to improve how models find, understand, and generate information. These days my research interests fall in three main categories, although I can get distracted by other LLM-based topics:
- Retrieval models: figuring out how to evaluate them comprehensively and giving them new capabilites, such as creating instructable/prompted retrievers
- Retrieval-Augmented Generation (RAG): working towards better RAG evaluations and improving RAG performance (often through better retrieval)
- Language model pre-training data: understanding why LMs act they way they do, curating corpora for pre-training or using pre-training information to help LM generation
Previously I graduated with my Bachelor’s degree from Brigham Young University in computer science and statistics, where I was advised by Kevin Seppi and Quinn Snell.
In the past, I’ve spent time interning with many excellent mentors: at Samaya AI in 2024 with Jack Hessel, Ashwin Paranjape, and Yuhao Zhang, at Semantic Scholar/AI2 working with Luca Soldaini, Kyle Lo, and Arman Cohan in 2023, at Apple AI/ML with Matthias Sperber in 2020 and 2021, and at AllenNLP/AI2 with Matt Gardner and Matthew Peters in 2020.
If you’re interested in getting in contact with me, please email me at {last_name}{first_name}@gmail.com.
news
Sep 2024 | Dated Data: Tracing Knowledge Cutoffs in Large Language Models accepted as a CoLM oral (top 10%) and a new preprint introducing the first promptable dense retrieval models! |
---|---|
Apr 2024 | Two new preprints and one accepted paper at SIGIR 2024: At SIGIR, On the Evaluation of Machine-Generated Reports and new preprints on evaluating and using instructions in IR and exploring LLM training data knowledge cutoffs and duplicates. |