BEGIN:VCALENDAR PRODID:-//eluceo/ical//2.0/EN VERSION:2.0 CALSCALE:GREGORIAN BEGIN:VEVENT UID:www.tcs.tifr.res.in/event/834 DTSTAMP:20230914T125940Z SUMMARY:Generative Adversarial Privacy: A Context-Aware Approach to Privacy -Guaranteed Data Publishing DESCRIPTION:Speaker: Lalitha Sankar (Arizona State University\nSchool of El ectrical\, Computer and Energy Engineering\n551 E. Tyler Mall\, Room 585\n Tempe\, AZ 85281\nUnited States of America)\n\nAbstract: \nPreserving the utility of published datasets while simultaneously providing provable priv acy guarantees is a well-known challenge. On the one hand\, context-free p rivacy solutions\, such as differential privacy\, provide strong privacy g uarantees\, but often lead to a significant reduction in utility. On the o ther hand\, context-aware privacy solutions\, such as information theoreti c privacy\, achieve an improved privacy-utility tradeoff\, but assume that the data curator has access to dataset statistics. We circumvent these li mitations by introducing a novel context-aware data-driven privacy framewo rk called generative adversarial privacy (GAP). GAP leverages recent advan cements in generative adversarial networks (GANs) to allow the data holder to learn privatization schemes from the dataset itself. Under GAP\, learn ing the privacy mechanism is formulated as a constrained minimax game betw een two players: a privatizer that sanitizes the dataset in a way that lim its the risk of inference attacks on the private variables\, and an advers ary that tries to infer the private variables from the sanitized dataset.   To evaluate the performance of GAP\, we investigate two simple (yet can onical) statistical dataset models: (a) the binary data model\, and (b) th e binary Gaussian mixture model. For both models\, we derive game-theoreti cally optimal minimax privacy mechanisms\, and show that the privacy mecha nisms learned from data (in a generative adversarial fashion) match the th eoretically optimal ones. This demonstrates that our framework can be easi ly applied in practice\, even in the absence of dataset statistics (joint work with Chong Huang (ASU)\, Peter Kairouz (Stanford)\, Xiao Chen (Stanfo rd)\, and Ram Rajagopal (Stanford).\nBio: Lalitha Sankar received the B.Te ch degree from the Indian Institute of Technology\, Bombay\, the M.S. degr ee from the University of Maryland\, and the Ph.D degree from Rutgers Univ ersity. She is presently an Assistant Professor in the ECEE department at Arizona State University. Prior to this\, she was an Associate Research Sc holar at Princeton University. Following her doctorate\, Dr Sankar was a r ecipient of a three-year Science and Technology teaching postdoctoral fell owship from the Council on Science and Technology at Princeton University. Her research interests include information privacy and cybersecurity in d istributed and cyber-physical systems. For her doctoral work\, she receive d the 2007-2008 Electrical Engineering Academic Achievement Award from Rut gers University. She received the IEEE Globecom 2011 Best Paper award for her work on side-information privacy and the US National Science Foundatio n CAREER award in 2014.\n URL:https://www.tcs.tifr.res.in/web/events/834 DTSTART;TZID=Asia/Kolkata:20171221T110000 DTEND;TZID=Asia/Kolkata:20171221T120000 LOCATION:A-201 (STCS Seminar Room) END:VEVENT END:VCALENDAR