Deep Learning Program Analysis: A New Paradigm for Analyzing Computer Programs

Benjamin Bales, from ASSRC, will give a talk on “Deep Learning Program Analysis: A New Paradigm for Analyzing Computer Programs”.

Date: April 1 2016, 11:30 AM
Location: Room B31, Inria Lille

Abstract: Since 2006, Deep Learning has outperformed many state of the art techniques on heterogeneous tasks, such as image recognition, machine translation, and voice recognition. The question remains as to whether or not it will prove effective for analyzing computer programs. In this talk, we discuss the current state of the art in deep learning and the initial approaches taken to apply neural architectures to computer source code. Specifically, we discuss some initial, and very promising, results for deep learning defect detection in computer source code achieved by QbitLogic, a start-up company based in Atlanta, GA (USA). We illuminate the challenges inherent to representing and training deep neural networks on software engineering data. We conclude with a call to action for the program analysis community to re-visit the challenging problems in areas such as defect detection and automated software repair armed with the recently developed deep learning toolkits of QbitLogic.