[NSF CAREER Award] NeuralSAT: A CDCL-based constraint solving approach to DNN Verification

Deep Neural Networks (DNNs)  are an incredibly powerful tool for tackling a wide-range of real-world problems, including image recognition, autonomous driving, power grid control, fake news detection, drug synthesis and discovery, and even COVID-19 detection and diagnosis. However, similar to any software system, DNNs can have "bugs" that cause unexpected results when presented with inputs that are different from those…