Federal Grant Award Summary

Oakland University received a $223,992 Project Grant from the National Science Foundation (NSF) under the Computer and Information Science and Engineering (CISE) program (CFDA 47.070), awarded June 1, 2026, with completion targeted for May 31, 2029. This collaborative research project focuses on developing fault localization techniques specifically designed for deep neural networks (DNNs), addressing a critical gap in software engineering practices. Traditional fault localization methods are inadequate for DNN models due to fundamental differences in computational architecture and the distinct nature of bugs in machine learning systems compared to conventional software.

The research will deliver novel approaches for monitoring and debugging DNN model behavior during neural network training, with particular emphasis on identifying dynamic behaviors that require instrumentation in execution traces. The project will explore three primary research directions: characterizing dynamic behaviors specific to different DNN architectures such as fully connected neural networks (FCNNs) and convolutional neural networks (CNNs); defining novel abstractions of these dynamic behaviors to enhance fault localization and repair capabilities; and developing practical tools to improve debugging accessibility for practitioners who lack deep expertise in DNN architecture. By enabling early error detection during the expensive training phase, these techniques have potential to reduce DNN development costs while improving the safety and reliability of AI-based software in mission-critical applications across healthcare, transportation, and other sectors.