Research on AI-assisted peer review, reviewer selection, scientific information retrieval, and computational discovery at the SOS+CD Lab.
Scientific knowledge is useful when people can find relevant work and evaluate it carefully. We study both the social processes of peer review and computational methods for navigating publications, figures, and research context.
Our review-generation work studies reasoning, reinforcement learning, and external context. The central questions concern the substance of a review and how well its comments are grounded in the manuscript. Evaluation scores should be interpreted within the tasks and models that produced them.
Peer review is also a social process. Our analysis of author-suggested reviewers examines how reviewer selection is associated with evaluations and acceptance outcomes, with attention to the limits of observational evidence.
Scientific documents contain more than prose. Models such as MISTI combine figures, captions, and publication metadata to improve retrieval. Related work in the archive addresses recommendations, datasets, and the interpretation of scholarly text.
How should review systems be evaluated when a fluent review, a high reward score, and a useful scientific criticism are different outcomes?
Explore information about joining the lab or contact Daniel Acuña about collaboration.
