Bridging the Expert Validation Gap in Biodiversity Monitoring with Inference-Aware Vendi Information Gain

2026 Faculty Research Award

AWARD PERIOD: 2026-2028

Monitoring how climate change and other environmental disruptions impact biodiversity is essential for conservation, but monitoring often generates enormous, unwieldy datasets that make timely conservation action difficult. Machine learning systems could make biodiversity monitoring easier and more efficient by automating many aspects, for example by identifying species from photos, videos, or audio recordings. However, ecologists currently lack a system for evaluating the quality of machine learning predictions’ ability to infer ecological patterns, such as changes in species distributions, population trends, and interactions between species. To fill this gap, Adji Bousso Dieng, Assistant Professor of Computer Science, and Lydia T. Liu, Assistant Professor of Computer Science, will develop a framework for evaluating the ability of machine learning predictions to inform ecological inferences. They will use this framework to test machine learning predictions based on video monitoring footage of hummingbird-flower interactions at the Rocky Mountain Biological Laboratory and acoustic recordings of bird calls from threatened biodiversity hotspots. By providing a framework for ecologists and conservation managers to more easily monitor biodiversity, this project will help accelerate biodiversity assessments and guide timely conservation action.