BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//UM//UM*Events//EN
CALSCALE:GREGORIAN
BEGIN:VTIMEZONE
TZID:America/Detroit
TZURL:http://tzurl.org/zoneinfo/America/Detroit
X-LIC-LOCATION:America/Detroit
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20070311T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20071104T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20250313T103532
DTSTART;TZID=America/Detroit:20250320T150000
DTEND;TZID=America/Detroit:20250320T170000
SUMMARY:Lecture / Discussion:2025 HFES Biomechanics Dinner Registration
DESCRIPTION:Dinner will be provided!\n\nAbout Dr. Pual Pridham:\n\nDr. Paul Pridham is a Lecturer and Research Area Specialist Senior in the Center for Ergonomics and Industrial and Operations Engineering Department at the University of Michigan. Paul completed a Ph.D. from Columbia University\, focused on rehabilitation robotics and how design impacts the performance of these systems. Paul went on to do postdocs at the Massachusetts Institute of Technology\, studying human exoskeleton fluency\; and at Northern Arizona University\, studying ankle exoskeletons with applications to clinical populations. Paul's work explores the impact of design on the performance of wearable systems\, particularly exoskeletons. This talk will examine how wearable systems can be used to better understand ourselves\, and in return how that understanding can allow us to better design wearable systems. Paul will also discuss future directions to take this work and encourage students to explore these topics.
UID:133815-21873591@events.umich.edu
URL:https://events.umich.edu/event/133815
CLASS:PUBLIC
STATUS:CONFIRMED
CATEGORIES:Dinner,Ergonomics,Human Factors And Ergonomics Society,Industrial And Operations Engineering,Michigan Engineering,North Campus
LOCATION:Pierpont Commons - Boulevard Room
CONTACT:
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20250313T093443
DTSTART;TZID=America/Detroit:20250320T150000
DTEND;TZID=America/Detroit:20250320T160000
SUMMARY:Workshop / Seminar:899 Seminar Series: Holden. Lee\, Johns Hopkins University
DESCRIPTION:Presenter Bio\nHolden Lee is an assistant professor of applied mathematics and statistics at Johns Hopkins University\, working on theoretical machine learning and applied probability. His research focuses on mathematical foundations for sampling algorithms and generative modeling\, with applications to language models. He was a co-organizer for the NeurIPS 2024 workshop on Creativity & Generative AI. He obtained his Ph.D. in mathematics from Princeton.\n\nAbstract\nMultimodal distributions such as mixture models pose significant challenges for both learning and sampling algorithms\, but score-based and diffusion models have enjoyed widespread empirical success on these problems. How can we understand their success from a theoretical perspective?\nFor learning\, we consider Gaussian mixtures and show diffusion models can learn a mixture of k isotropic Gaussians in R^n with quasi-polynomial time and sample complexity (exp(poly log((n+k)/epsilon)))\, under a minimum weight assumption. This gives a completely different\, analytic proof of a result previously known only using a specialized algebraic approach\, and moreover\, it extends to give the first efficient algorithm for learning a non-parametric family of Gaussian convolutions of distribution supported on sets with a small covering number.\n\nFor sampling\, we explain the surprising observation that despite the slow mixing of Langevin dynamics\, it is possible to sample from multimodal distributions by learning only the vanilla score\, as long as we use data-based initialization. We consider the more general problem of sampling using a Markov chain without global mixing given a small number of samples from the stationary measure\, showing efficient sampling with a number of data samples almost linear in the number of modes\; this also covers the case of Glauber dynamics with pseudolikelihood estimation.\nBased on joint work with Khashayar Gatmiry and Jonathan Kelner (MIT)\, Frederic Koehler (UChicago)\, Thuy-Duong Vuong (UC Berkeley) on the papers https://arxiv.org/abs/2404.18869 and https://arxiv.org/abs/2411.09117.\n\nThis event is a part of our 899 seminar series.
UID:133798-21873576@events.umich.edu
URL:https://events.umich.edu/event/133798
CLASS:PUBLIC
STATUS:CONFIRMED
CATEGORIES:899 Seminar Series,Industrial And Operations Engineering,seminar
LOCATION:Industrial and Operations Engineering Building - G690
CONTACT:
END:VEVENT
END:VCALENDAR