Computational research on scientific image integrity, responsible AI, reporting quality, and the availability of research resources.
Research integrity depends on evidence that people can inspect. We develop computational methods to identify patterns that deserve attention, measure uncertainty, and help researchers evaluate scientific work.
Our work examines image reuse, image manipulation, and graphical inconsistencies. A detection method helps locate and characterize a problem; interpretation still depends on the original material and the scientific context.
Language models can assist scientific work, but their behavior needs evaluation. We study how models respond to research-integrity scenarios, including requests that disguise misconduct as ordinary scientific tasks.
Links to code and data are part of the evidence behind a paper. Our resource-longevity work examines why these links persist or disappear, while journal-screening research investigates the signals and limits of automated assessment.
How can screening systems communicate uncertainty clearly enough to help investigators without turning a model's score into a verdict?
Explore information about joining the lab or contact Daniel Acuña about collaboration.
