In many real-world problems, a big scale gap can be observed between micro- and macroscopic scales of the problem because of the difference in mathematical (engineering, social, biological, physical, etc.) models and/or laws at different scales. The main objective of multiscale algorithms is to create a hierarchy of problems, each representing the original problem at different coarse scales with fewer degrees of freedom. We will discuss the multiscale frameworks for (nonlinear) support vector machines, and several optimization and mining problems on graphs along with their application on the near-term quantum devices (including quantum universal and annealing machines). We will present a scalable multilevel framework for SVM that is based on the elements of algebraic multigrid and demonstrate a substantial improvement of the model training computational time, and other advantages.
Short bio: Dr. Ilya Safro is an Associate Professor in the Department of Computer and Information Sciences. Before joining UD he was an Associate Professor in the School of Computing and Faculty Scholar in the School of Health Research at Clemson University, where he was a director of the Algorithms and Computational Science Lab. Dr. Safro received his PhD in Applied Mathematics and Computer Science in the Weizmann Institute of Science. In 2008-2012, Dr. Safro was a postdoc and Argonne scholar at Argonne National Laboratory with a joint appointment in the University of Chicago. His research interests include algorithms in machine learning, AI, NLP, network science, and quantum computing. Dr. Safro is an academic editor in the machine learning and network science areas of PLOS One and Algorithms. He is also a member of advisory boards of several companies. He has organized several week-long workshops on high-performance graph algorithms, tutorials and mini-symposia at SIAM conferences.