Federal Grant Award Summary

The National Science Foundation's Division of Computing and Communication Foundations awarded a CAREER grant of $371,191 to Boston University (awarded September 1, 2026; completion August 31, 2031) under the Computer and Information Science and Engineering program (CFDA 47.070). This project delivers a developmentally plausible, data-efficient pretraining framework for Vision Foundation Models (VFMs)—artificial intelligence systems designed for comprehensive image and video understanding. The primary deliverable is a training methodology inspired by human infant cognitive development that significantly reduces the computational and financial burden of creating VFMs. By leveraging longitudinal, egocentric audiovisual recordings of infants, the research team will establish a core framework, design evaluation benchmarks aligned with early cognitive milestones, and develop methods to bridge sensory and temporal gaps in training data through model ensembling techniques.

The project's products include curated datasets of infant egocentric recordings, a formally defined pretraining framework, and associated evaluation benchmarks that make VFM development accessible to university-scale research budgets rather than requiring the substantial resources of highly funded institutions. By democratizing Vision Foundation Model research, this work aims to increase transparency, advance artificial intelligence safety, and build public trust in AI systems. The research also generates transferable insights applicable to specialized domains such as medical imaging and vocational training where data availability is limited, while providing educational opportunities for students and expanding community participation in foundational AI model development.