BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//7.2.3.1//EN
TZID:America/New_York
X-WR-TIMEZONE:America/New_York
BEGIN:VEVENT
UID:829@fds.yale.edu
DTSTART;TZID=America/New_York:20250226T113000
DTEND;TZID=America/New_York:20250226T130000
DTSTAMP:20250916T142147Z
URL:https://fds.yale.edu/events/fds-colloquium-brice-huang-mit/
SUMMARY:FDS Colloquium: Brice Huang (MIT)\, "Algorithmic thresholds in rand
 om optimization problems"
DESCRIPTION:\nAbstract: Optimizing high-dimensional functions generated fro
 m random data is a central problem in modern statistics and machine learni
 ng. As these objectives are highly non-convex\, the maximum value reachabl
 e by efficient algorithms is usually smaller than the maximum value that e
 xists\, and characterizing the fundamental computational limits of these p
 roblems is a difficult challenge.\n\n\n\nIn this talk\, I will describe th
 e "branching overlap gap property\," a new technique I have developed that
  obtains sharp algorithmic thresholds in several random optimization probl
 ems. We focus on two prototypical families of random functions\, namely th
 e Hamiltonians of mean-field spin glasses and random perceptrons. These mo
 dels are widely studied in probability\, statistical physics\, and compute
 r science\, and are related to problems in high-dimensional statistical in
 ference and neural networks. We exactly characterize the maximum value ach
 ievable by a broad class of stable algorithms\, which includes gradient de
 scent\, Langevin dynamics\, and general first-order methods on dimension-f
 ree time scales. We also identify this value with a phase transition in th
 e solution landscape\, yielding a unified geometric description of the alg
 orithmic threshold that holds across several problems.\n\n\n\nSpeaker bio:
  Brice Huang is a final-year Ph.D. student in Electrical Engineering and C
 omputer Science at MIT\,  advised by Nike Sun and Guy Bresler. His researc
 h focuses on the theoretical foundations of optimization and sampling prob
 lems over random\, high-dimensional landscapes\, and studies both the fund
 amental statistical and computational limits of these problems. In additio
 n\, he has worked on statistical inference\, quantum learning theory\, and
  combinatorics. His work has been recognized with a FOCS Best Student Pape
 r Award\, the Ernst A. Guillemin award for Best SM thesis\, and a Google P
 h.D. Fellowship.\n
CATEGORIES:FDS Events
LOCATION:Yale Institute for Foundations of Data Science\, Kline Tower 13th 
 Floor\, Room 1327\, New Haven\, CT\, 06511\, United States
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Kline Tower 13th Floor\, Ro
 om 1327\, New Haven\, CT\, 06511\, United States;X-APPLE-RADIUS=100;X-TITL
 E=Yale Institute for Foundations of Data Science:geo:0,0
END:VEVENT
BEGIN:VTIMEZONE
TZID:America/New_York
X-LIC-LOCATION:America/New_York
BEGIN:STANDARD
DTSTART:20241103T010000
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
END:STANDARD
END:VTIMEZONE
END:VCALENDAR