PhD student position at the School of Engineering Sciences, KTH Royal Institute of Technology.
Project description
We focus on developing optimization algorithms, methods, and fundamental theory, for analyzing specific input-output properties of deep neural networks (DNNs) and training DNNs with guaranteed properties. The ability to rigorously analyze and train DNNs with guaranteed properties is essential for us to trust AI-based systems, and this is an active research area. The goal is to develop deterministic optimization methods and we will build upon, and further develop, theory and concepts from mixed-integer and global optimization. A central question in the project is ”how to construct strong polyhedral representations and relaxations of DNNs”. A strong interest in mathematical programming (e.g., mixed-integer optimization) is a great advantage.