Workshop’s Topic: In the classical statistical theory of sequential prediction, the sequence of elements, which we call outcomes, is assumed to be a realization of a stationary stochastic process. Under this hypothesis, statistical properties of the process may be estimated on the basis of the sequence of past observations, and effective prediction rules can be derived from these estimates. In such a setup, the risk of a prediction rule may be defined as the expected value of some loss function measuring the discrepancy between predicted value and true outcome and different rules are compared based on the behavior of their risk.

In this short course, we look at prediction from a quite different angle. We abandon the basic assumption that the outcomes are generated by an underlying stochastic process and view the sequence as the product of some unknown and unspecified mechanism (which could be deterministic, stochastic, or even adversarially adaptive to our own behavior). We discuss the fundamental ideas including both algorithms and analyses - in this framework through the unified lens of learning from experts’ advice.

Time and Location: 14:00-17:00, Aug 12 (Peer Review) | 9:00-12:00/14:00-17:00, Aug 13-15 (Keynote Speech), Room C110 (School of Management)

Language: Bilingual (Chinese and English)