BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Alexandre d'Aspremont (École Normale Supérieure Paris (ENS)) DTSTART:20200615T130000Z DTEND:20200615T140000Z DTSTAMP:20260423T035024Z UID:OWOS/2 DESCRIPTION:Title: Nai ve feature selection: Sparsity in naive Bayes\nby Alexandre d'Aspremon t (École Normale Supérieure Paris (ENS)) as part of One World Optimizati on seminar\n\n\nAbstract\nDue to its linear complexity\, naive Bayes class ification remains an attractive supervised learning method\, especially in very large-scale settings. We propose a sparse version of naive Bayes\, w hich can be used for feature selection. This leads to a combinatorial maxi mum-likelihood problem\, for which we provide an exact solution in the cas e of binary data\, or a bound in the multinomial case. We prove that our b ound becomes tight as the marginal contribution of additional features dec reases. Both binary and multinomial sparse models are solvable in time alm ost linear in problem size\, representing a very small extra relative cost compared to the classical naive Bayes. Numerical experiments on text data show that the naive Bayes feature selection method is as statistically ef fective as state-of-the-art feature selection methods such as recursive fe ature elimination\, l1-penalized logistic regression and LASSO\, while bei ng orders of magnitude faster. For a large data set\, having more than wit h 1.6 million training points and about 12 million features\, and with a n on-optimized CPU implementation\, our sparse naive Bayes model can be trai ned in less than 15 seconds.\n\nThe talk is based on joint work with Armin Askari and Laurent El Ghaoui that can be found at https://arxiv.org/abs/1 905.09884\n\nthe address and password of the zoom room of the seminar are sent by e-mail on the mailinglist of the seminar one day before each talk\ n LOCATION:https://researchseminars.org/talk/OWOS/2/ END:VEVENT END:VCALENDAR