Abstract: Despite the overwhelming success of neural networks for pattern recognition, these models behave categorically different from humans. Adversarial examples, small perturbations which are often undetectable to the human eye, easily fool neural networks, demonstrating that neural networks lack the robustness of human classifiers. This defense comprises two parts. First, we develop methods for hardening neural networks against an adversary. Second, we discuss several mathematical properties of the neural network models we use. These properties are of interest beyond robustness to adversarial examples, and they extend to the broad setting of deep learning.
Dissertation directed by: Wojciech Czaja
In view of recent developments on our campus, the defense will be held virtually on Zoom. The virtual meeting will start at 9:50, and the defense will begin at 10:00 AM on March 30, 2020. As this is new for most of us, I ask you to join during this 10 minute period before 10 AM.
To attend the virtual defense, please join the Zoom session at the following link: https://umd.zoom.us/j/834093556 (For regular Zoom users, the meeting ID is: 834 093 556.)
Zoom sessions can be accessed in the browser or by downloading the Zoom app. UMD students and faculty can access and join Zoom at umd.zoom.us by using your UMD credentials. Please note that not all web browsers work well with Zoom. Therefore downloading the app is encouraged, esp., for habitual Firefox users.
Abstract: In this talk, we introduce methods from convex optimization to solve the multi-marginal transport-type problems that arise in the context of density functional theory. Convex relaxations are used to provide outer approximation to the set of N-representable 2-marginals and 3-marginals, which in turn provide lower bounds to the energy. We further propose rounding schemes to obtain upper bounds to the energy.
Abstract: This talk is about some recent progress on solving inverse problems using deep learning. Compared to traditional machine learning problems, inverse problems are often limited by the size of the training data set. We show how to overcome this issue by incorporating mathematical analysis and physics into the design of neural network architectures. We first describe neural network representations of pseudodifferential operators and Fourier integral operators. We then continue to discuss applications including electric impedance tomography, optical tomography, inverse acoustic/EM scattering, seismic imaging, and travel-time tomography.
Abstract: This talk is about some recent progress on solving inverse problems using deep learning. Compared to traditional machine learning problems, inverse problems are often limited by the size of the training data set. We show how to overcome this issue by incorporating mathematical analysis and physics into the design of neural network architectures. We first describe neural network representations of pseudodifferential operators and Fourier integral operators. We then continue to discuss applications including electric impedance tomography, optical tomography, inverse acoustic/EM scattering, seismic imaging, and travel-time tomography.