BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Batya Kenig (Technion) DTSTART:20231108T160000Z DTEND:20231108T171500Z DTSTAMP:20260423T035956Z UID:AAIT/24 DESCRIPTION:Title: Ap proximate Implication for Probabilistic Graphical Models.\nby Batya Ke nig (Technion) as part of Seminar on Algorithmic Aspects of Information Th eory\n\n\nAbstract\nThe graphical structure of Probabilistic Graphical Mod els (PGMs) represents the conditional independence (CI) relations that hol d in the modeled distribution. The premise of all current systems-of-infer ence for deriving conditional independence relations in PGMs\, is that the set of CIs used for the construction of the PGM hold exactly. In practice \, algorithms for extracting the structure of PGMs from data discover appr oximate CIs that do not hold exactly in the distribution. In this work\, w e ask how the error in this set propagates to the inferred CIs read off th e graphical structure. More precisely\, what guarantee can we provide on t he inferred CI when the set of CIs that entailed it hold only approximatel y? In this talk\, I will describe new positive and negative results concer ning this problem. \n\nBased on:\nhttps://lmcs.episciences.org/8943 \;\nht tps://proceedings.mlr.press/v161/kenig21a.html \;\nhttps://arxiv.org/abs/2 310.13942\n LOCATION:https://researchseminars.org/talk/AAIT/24/ END:VEVENT END:VCALENDAR