Abstract
In this talk, we introduce alpha-loss as a parameterized class of loss functions that resulted from operationally motivating information-theoretic measures. Tuning the parameter alpha from 0 to infinity yields a class of loss functions that admit continuous interpolation between log-loss (alpha=1), exponential loss (alpha=1/2), and 0-1 loss (alpha=infinity). We discuss how different regimes of the parameter alpha enables the practitioner to tune the sensitivity of their algorithm towards two emerging challenges in learning: robustness and fairness. We discuss classification properties of the class, information-theoretic interpretations, and the optimization landscape of the average loss as viewed through the lens of Strict-Local-Quasi-Convexity under the logistic regression model. Finally, we comment on ongoing and future work on different applications of alpha-loss including deep neural networks, federated learning, and boosting.
Joint work with Tyler Sypherd, Mario Diaz, Gautam Dasarathy, Peter Kairouz, and John Cava
Biography
Lalitha Sankar is an Associate Professor in the School of Electrical, Computer, and Energy Engineering at Arizona State University where she was an Assistant Professor from 2012 to 2018. Prior to that she was a Research Scholar in the Department of Electrical Engineering at Princeton University working with H. Vincent Poor. She was also a Science and Technology Teaching and Research Fellow supported by the Council on Science and Technology at Princeton University. Sankar obtained her PhD from Rutgers University, MS from the University of Maryland and a bachelor’s degree from the Indian Institute of Technology, Bombay, India.
Lalitha Sankar received the best paper award from the IEEE Globecom 2011 for her paper on side-information privacy with R. Tandon and H. V. Poor. For her doctoral work, she received the 2007-2008 Electrical Engineering Academic Achievement Award from Rutgers University. She received the NSF CAREER award in 2014. She leads an NSF HDR Institute focused on developing data science that enables integration of high-dimensional spatio-temporal synchrophasor data into electric grid operations.