BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Vladimir Vovk (University of London\, U.K.) DTSTART:20210315T103000Z DTEND:20210315T111500Z DTSTAMP:20260421T174009Z UID:BPS/24 DESCRIPTION:Title: Mul tiple hypothesis testing with e-values\nby Vladimir Vovk (University o f London\, U.K.) as part of Bangalore Probability Seminar\n\n\nAbstract\nh ttps://www.isibang.ac.in/~statmath/pcm2020/\n\nMartingales in the foundati ons of statistics\nDate: March 15th\, 2021\nTime: 15:00-15:45 IST\n\n\nAbs tract: Traditional methods of testing statistical hypotheses have been dev eloped for the batch setting\, in the terminology of machine learning: giv en a batch of data\, statisticians typically compute measures of disagreem ent\, such as p-values or Bayes factors\, between a null hypothesis and th e data. An alternative that is popular in machine learning is the online s etting\, in which the items of data (observations ) keep arriving sequenti ally. In this introductory lecture I will explain the role of martingales\ , in the form of test martingales\, in online hypothesis testing and discu ss their applications in the foundations of probability and statistics.\n\ nMultiple hypothesis testing with e-values\nDate: March 15th\, 2021\nTime: 16:00-16:45 IST\n\nAbstract: It is interesting that test martingales do n ot trivialize in the case of only one observation. In fact\, they provide a useful alternative\, sometimes called e-values\, to the standard statist ical notion of p-values. The most important mathematical advantage of e-va lues over p-values is that the average of e-values is always an e-value. T his property is valuable in multiple hypothesis testing\, which will be th e topic of this lecture.\n\nConformal prediction\nDate: March 19th\, 2021\ nTime: 15:00-15:45 IST\n\nAbstract: Mainstream machine learning\, despite its recent successes\, has a serious drawback: while its state-of-the-art algorithms often produce excellent predictions\, they do not provide measu res of their accuracy and reliability that would be both practically usefu l and provably valid. On the other hand\, such measures are commonplace in statistics. Conformal prediction adapts rank tests\, popular in nonparame tric statistics\, to testing the IID assumption (the observations being in dependent and identically distributed)\, which is the standard assumption made in machine learning. This gives us practical measures\, provably vali d under the IID assumption\, of the accuracy and reliability of prediction s produced by traditional and recent machine-learning algorithms. In this lecture I will give a brief review of conformal prediction.\nConformal hyp othesis testing\nDate: March 19th\, 2021\nTime: 16:00-16:45 IST\n\nAbstrac t: An interesting application of conformal prediction is the existence of exchangeability martingales\, i.e.\, random processes that are test martin gales under any exchangeable probability measure. In particular\, they are martingales whenever the observations are IID. The topics of this last le cture in this series will be the construction of exchangeability martingal es and their use for different kinds of change detection\, including detec ting a point at which the IID assumption becomes violated and detecting co ncept shift.\n LOCATION:https://researchseminars.org/talk/BPS/24/ END:VEVENT END:VCALENDAR