Project Grant 2205004

This Project Grant from the National Science Foundation's Computer and Information Science and Engineering program (CFDA 47.070) provides $600,000 to The Pennsylvania State University from October 1, 2022 to September 30, 2025. The university will develop interpretable machine learning methods based on deep neural networks from a source coding perspective. Researchers will draw an analogy between explaining complex prediction models and transmitting signals with limited channel capacity. The goal is to quantize prediction mappings to allow descriptions at desired interpretability levels, using a mixture of discriminative models trained as an embedded part of a neural network. Function approximation theory for interpretable neural networks will also be developed. Researchers will test the framework on benchmark datasets covering images, video, text and biomedical data, with a focus on emotion recognition from body movements. Collaboration with Texas A&M University will support this work. Outcomes include software, trained students, and integrated teaching materials applying the research.