<html><head><meta http-equiv="content-type" content="text/html; charset=utf-8"></head><body dir="auto"><span style="background-color: rgba(255, 255, 255, 0);">Departments of Mathematics & Computer Science<br>Combinatorics & Theory Seminar<br><br><a href="x-apple-data-detectors://0" dir="ltr" x-apple-data-detectors="true" x-apple-data-detectors-type="calendar-event" x-apple-data-detectors-result="0" style="text-decoration-color: rgba(0, 0, 0, 0.258824);">Tuesday, November 5</a><br>Ry 251 <a href="x-apple-data-detectors://1" dir="ltr" x-apple-data-detectors="true" x-apple-data-detectors-type="calendar-event" x-apple-data-detectors-result="1" style="text-decoration-color: rgba(0, 0, 0, 0.258824);">@3:30pm</a><br><br>Aleksandar Nikolov<br>University of Toronto<br><br>Title: The Power of Factorization<br><br>Abstract: A central goal in private data analysis is to estimate statistics about an unknown distribution from a dataset possibly containing sensitive information, so that the privacy of any individual represented in the dataset is preserved. We study this question in the model of non-interactive local differential privacy (LDP), in which every person in the dataset randomizes their own data in order to preserve its privacy, before sending it to a central server. We give a characterization of the minimum number of samples necessary to get an accurate estimates of a given set of statistical queries, as well as a characterization of the sample complexity of agnostic PAC learning in this model. The characterization is tight up polylogarithmic factors for any given set of statistical queries, and, respectively, any given concept class. The characterization is achieved by a simple and efficient instance-optimal (with respect to the queries/concept class) approximate factorization mechanism, i.e. a mechanism that answers the statistical queries by answering a different set of strategy queries, from which the answers to the original queries can be approximately reconstructed.<br><br>Based on joint work with Alexander Edmonds and Jonathan Ullman<br><br><br>Refreshments will be served prior to the talk <a href="x-apple-data-detectors://3" dir="ltr" x-apple-data-detectors="true" x-apple-data-detectors-type="calendar-event" x-apple-data-detectors-result="3" style="text-decoration-color: rgba(0, 0, 0, 0.258824);">at 3:15 pm</a> Ry 257<br></span><div dir="ltr"></div></body></html>