BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Brian Nord (Fermilab) DTSTART:20210504T170000Z DTEND:20210504T180000Z DTSTAMP:20260423T024534Z UID:nhetc/13 DESCRIPTION:Title: D eeply Uncertain: (how) can we make deep learning tools trustworthy for sci entific measurements?\nby Brian Nord (Fermilab) as part of NHETC Semin ar\n\n\nAbstract\nArtificial Intelligence (AI) --- including machine learn ing and deep learning --- refers to a set of techniques that rely primaril y on the data itself for the construction of a quantitative model. AI has been in development for about three quarters of a century\, but there has been a recent resurgence in research and applications. This current (third ) wave of AI progress is marked by extraordinary results --- for example\, in image analysis\, language translation\, and machine automation. Despit e the modest definition of AI\, its potential to disrupt technologies\, ec onomies\, society\, and even science is often presented as unmatched in mo dern times. However\, along with the promise of AI\, there are significant challenges to overcome to reach a degree of reliability that is on par wi th more traditional modeling methods. \nIn particular\, uncertainty quanti fication metrics derived from deep neural networks are yet to be made phys ically interpretable. For example\, when one uses a convolutional neural n etwork to measure values from an image (e.g.\, regression for galaxy prope rties)\, the error estimates do not necessarily match those from an MCMC l ikelihood fit. In this presentation\, I will discuss the landscape of unce rtainty quantification in deep learning\, as well as some computational ex periments in a physical context that demonstrate a mismatch between errors derived directly from deep learning methods and those derived through tra ditional error propagation. Before we can apply deep learning tools confid ently for the direct measurement of physical properties\, we’ll need sta tistically robust error estimation methods.\n LOCATION:https://researchseminars.org/talk/nhetc/13/ END:VEVENT END:VCALENDAR