Tuesday 12pm, 30 April 2019
Learning to Computationally Design Engineered Systems
Assistant Professor - University of Maryland, College Park
This talk will highlight our recent work on combining three scientific fields that may, at first glance, appear tangential to one another—machine learning, engineering systems design, and multi-physics simulation—in order to automatically design and explore new complex engineered systems ranging in size from atomic scale condensed matter physics/chemistry to systems the size of aircraft and cargo ships and even up to globally distributed communities of thousands of people collaborating together. As we briefly walk through each of these size scales and see various physical phenomena emerge—e.g., how materials pack into 3D space, how modifications to a surface of an airfoil affect fuel efficiency, how engineers collaborate together on global scales, etc.—we also see how each possesses unique mathematical structures that occur in or can be leveraged by Machine Learning—e.g., Differential Geometry, Manifold subspaces, Submodular Functions, etc. I will discuss how those structures help us understand how humans design engineered systems and how that knowledge can help us better teach computers how to emulate that process. As a byproduct, this connection also points the way to advances in fundamental problems in computer science, like Diverse Bipartite Matching, Diverse Ranking, clustering Positive Semi-Definite Matrices on Riemannian Manifolds, and high-dimensional non-convex optimization, among other topics.
Overall, this talk's central argument is that Machine Learning and Engineering Systems have a powerful symbiotic relationship: (1) that Machine Learning is not only a valuable tool for helping understand real-world engineering systems but also (2) that Engineering itself helps shed light on phenomena with interesting and unique mathematical structure that can inform new algorithmic advances within ML not well covered by existing ML benchmarks. Both are trying, in complementary ways, to uncover fundamental representations of nature, of complexity, and of human behavior. I argue (and attempt to briefly show) how the two fields are better together than they are apart, and that their merger provides a unique window into understanding fundamental questions of nature and human-kind in ways that are not possible otherwise.
Mark Fuge is an Assistant Professor of Mechanical Engineering at the University of Maryland, College Park, where he is also an affiliate faculty in the Institute for Systems Research and a member of the Maryland Robotics Center and Human-Computer Interaction Lab. His staff and students study fundamental scientific and mathematical questions behind how humans and computers can work together to design better complex engineered systems, from the molecular scale all the way to systems as large as aircraft and ships, by using tools from Applied Mathematics (such as graph theory, category theory, and statistics) and Computer Science (such as machine learning, artificial intelligence, complexity theory, and submodular optimization). He received his Ph.D. from UC Berkeley and has received a DARPA Young Faculty Award, a National Defense Science and Engineering Graduate (NDSEG) Fellowship, and has prior/current support from NSF, NIH, DARPA, ONR, and Lockheed Martin. You can learn more about his research at his lab’s website: http://ideal.umd.edu