Federal Grant Award Summary

Purdue University received a $339,075 CAREER Project Grant from the National Science Foundation's Division of Computing and Communication Foundations under the Computer and Information Science and Engineering (CISE) program (CFDA 47.070), effective August 1, 2026 through July 31, 2031. The award funds research and development of data-centric vision models that enhance the interpretability, controllability, and maintainability of artificial intelligence systems used in computer vision applications. The project addresses limitations in current fully parametric deep learning models by developing semi-parametric architectures that combine the performance strengths of parametric models with the transparency and explainability of non-parametric methods, enabling AI predictions to be directly traced to specific training examples.

The research is organized around four primary technical thrusts: developing semi-parametric architectural frameworks, and applying them to the problems of machine unlearning, data attribution, and model customization. By explicitly incorporating training data at inference time, these models will enable more transparent diagnosis of undesirable outputs, adaptation to shifting data distributions, and attribution of specific model behaviors to their underlying training data sources. This approach directly addresses practical deployment requirements for safe, accountable, and interpretable AI systems as artificial intelligence becomes more embedded in everyday applications.