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:20260126T132912
DTSTART;TZID=America/Detroit:20260212T150000
DTEND;TZID=America/Detroit:20260212T160000
SUMMARY:Lecture / Discussion:IOE 899: Soroosh Shafiee
DESCRIPTION:We investigate the problem of certifying optimality for sparse generalized linear models (GLMs)\, where sparsity is enforced through a cardinality constraint. While Branch-and-Bound (BnB) frameworks can certify optimality using perspective relaxations\, existing methods for solving these relaxations are computationally intensive\, limiting their scalability. To address this challenge\, we propose a unified proximal first-order algorithmic framework that is both linearly convergent and computationally efficient. We first develop a general theory for composite optimization problems satisfying specific geometric regularity conditions. \n\nBy establishing a rigorous link between primal quadratic growth and dual quadratic decay\, we derive novel error bounds showing that the computable duality gap can serve as a tight proxy for the distance to the solution set. Leveraging this property\, we design a restart scheme that upgrades generic sublinear algorithms to achieve provable linear convergence for both primal and dual objectives. We then instantiate this framework for the perspective relaxation of sparse GLMs. \n\nWe prove that standard GLM loss functions and the implicit perspective regularizer satisfy the required geometric conditions. Furthermore\, we develop specialized algorithms to evaluate the non-smooth regularizer and its proximal operator exactly in log-linear time\, avoiding the high cost of generic conic solvers. Extensive experiments on synthetic and real-world datasets demonstrate that our approach leverages GPU acceleration to speed up dual bound computations by orders of magnitude\, significantly enhancing the capability of BnB to certify optimality for large-scale problems.
UID:144573-21895503@events.umich.edu
URL:https://events.umich.edu/event/144573
CLASS:PUBLIC
STATUS:CONFIRMED
CATEGORIES:899 Seminar Series,Graduate,Graduate Students,Industrial And Operations Engineering,Michigan Engineering
LOCATION:Industrial and Operations Engineering Building - 1680
CONTACT:
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20260202T094257
DTSTART;TZID=America/Detroit:20260212T153000
DTEND;TZID=America/Detroit:20260212T162000
SUMMARY:Lecture / Discussion:The Department of Astronomy 2025-2026 Colloquium Series Presents:
DESCRIPTION:\"Shooting for the Stars: Jet-Wind Interactions and their Role in Understanding Jet Physics\"\n\nRelativistic Jets carry vast amounts of energy and magnetic flux from accreting black holes into the interstellar and intergalactic medium. They are routinely invoked as key sources of feedback in structure formation. And yet\, after over 60 years of study\, the mechanisms through which they couple with their environment\, and even their fundamental properties like particle content\, velocity\, and instantaneous power are still poorly understood. I will discuss a mechanism by which we can shed light on some of the outstanding questions in jet physics: Their interaction with stellar winds\, on scales from Galactic X-ray binaries to supermassive black holes in galaxy clusters.
UID:144903-21896126@events.umich.edu
URL:https://events.umich.edu/event/144903
CLASS:PUBLIC
STATUS:CONFIRMED
CATEGORIES:astronomy,astrophysics
LOCATION:West Hall - 411
CONTACT:
END:VEVENT
END:VCALENDAR