Meeting ID: 926 1895 4954
+13017158592,,92618954954# US (Germantown)
+13126266799,,92618954954# US (Chicago)
Thomas Goldstein, PhD
University of Maryland, Department of Computer Science
?Evasionand poisoning attacks on neuralnetworks: theoretical andpractical perspectives?
Abstract: In this talk I will give anoverview of adversarial attacks and dataset poisoning attacks on neuralnetworks. Using empirical studies, I will showexamples where these attacks can pose a real threat to real-worldsystems, such as copyright detection system, financial markets, andGoogle’s AutoMLAPI. Then, I’ll dive into the theory of adversarial attacks, and presentsituations where such attacks cannot be avoided.
Bio: Thomas Goldstein obtained his PhDin Mathematics at UCLA, and was a research scientist atRice University andStanford University. He has been the recipient of several awards, including SIAM’sDiPrimaPrize,a DARPA Young Faculty Award, and a SloanFellowship.His research lies at the intersection of machine learning and optimization, andtargets applications in computer vision and signal processing. Dr,Goldstein works at the boundary between theory and practice, leveragingmathematical foundations, complex models, and efficient hardware to buildpractical, high-performance systems. He designs optimization methods for a widerange of platforms ranging from powerful cluster/cloud computing environmentsto resource limited integrated circuits and FPGAs.