
"Readying Machine Learning for Quantum Computing"
Fri, May 27, 2011 @ 02:00 PM  03:00 PM
Ming Hsieh Department of Electrical and Computer Engineering
Conferences, Lectures, & Seminars
Speaker: Hartmut Neven, Google
Talk Title: "Readying Machine Learning for Quantum Computing"
Abstract: Modern approaches to machine learning formulate training of a classifier as an optimization problem in which simultaneously the training error as well as the classifier complexity is minimized. For computational efficiency typically a convex objective is constructed. But it is well known that such a choice comes at a cost. For instance, convex loss functions designed to measure training performance are not as robust to noise as their nonconvex counter parts and convex regularization does
not achieve as high levels of sparsity as versions involving the L0norm. Nonconvex losses also figure prominently in recent attempts to derive tighter bounds for the generalization error. Here we report on experiments to train with nonconvex objectives using discrete optimization in a formulation adapted to take advantage of emerging hardware for quantum optimization. A key finding is that the resulting
classifiers are already competitive when using as temporary standin a classical heuristic solver. We will give an overview of the state of the quantum hardware development as well as what advantages in terms of quality of the solution we can hope to attain from a theoretical point of view.
Host: Daniel Lidar
Location: Seaver Science Library (SSL)  150
Audiences: Everyone Is Invited
Contact: Daniel Lidar