BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:America/Chicago BEGIN:DAYLIGHT DTSTART:20260308T030000 TZOFFSETFROM:-0600 TZOFFSETTO:-0500 RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3 TZNAME:CDT END:DAYLIGHT BEGIN:STANDARD DTSTART:20261101T010000 TZOFFSETFROM:-0500 TZOFFSETTO:-0600 RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11 TZNAME:CST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20260905T221646Z UID:B943DE69-F3F5-4CCD-8C0C-C0EA00FDA0EA DTSTART;TZID=America/Chicago:20260918T110000 DTEND;TZID=America/Chicago:20260918T121500 DESCRIPTION:IEEE Information Theory Society Distinguished Lecture\n\nTitle: Generative Modeling in Probability Space: From Wasserstein Optimization t o Local Flow Matching\n\nSpeaker:Yao Xie ([Official webpage](https://www.i sye.gatech.edu/users/yao-xie)) ([CV](https://www2.isye.gatech.edu/~yxie77/ YaoXie_CV.pdf)) ([2-page CV](https://www2.isye.gatech.edu/~yxie77/YaoXie_2 page.pdf))\n[Coca-Cola Foundation Chair](https://www.isye.gatech.edu/news/ yao-xie-driving-data-science-innovation-coca-cola-foundation-chair) and Pr ofessor\n[H. Milton Stewart School of Industrial and Systems Engineering ( ISyE)](http://www.isye.gatech.edu/)\nAssociate Director\, [Machine Learnin g Center](https://ml.gatech.edu/)\nAdjunct Professor\, [School of Electric al and Computer Engineering (ECE)](https://www.ece.gatech.edu/)\n[Georgia Institute of Technology](http://www.gatech.edu/)\n\nAbstract: How can we l earn complex probability distributions through a sequence of simpler trans formations with theoretical guarantees? This talk presents flow-based gene rative modeling as iterative computation in probability space\, connecting optimization\, transport\, and diffusion. I will first discuss JKO flow m odels\, which implement proximal optimization in Wasserstein space. I will explain how convergence of the forward data-to-noise process yields guara ntees for reverse generation\, with bounds in Kullback-Leibler divergence that account for per-step optimization errors. I will then introduce Local Flow Matching\, which learns short transformations between nearby distrib utions along a diffusion path. This construction enables smaller models an d efficient training\, with generation guarantees in chi-square divergence derived from diffusion contraction. Together\, these results illustrate h ow optimization in probability space can guide the design and analysis of generative models. The talk will emphasize mathematical intuition and sele cted theoretical results\, complemented by numerical examples. This is bas ed on joint work with Xiuyuan Cheng\, Jianfeng Lu\, Yixin Tan\, and Chen X u.\n\nDate: September 18\, 2026\, from 11:00am - 12:15pm Chicago time\n\nL ocation: UIC east campus\, lecture center C1\, Chicago\, IL 60607\n\nSpeak er(s): \, Yao\n\nRoom: Lecture Center C1\, Bldg: Lecture Center C1\, UIC e ast campus\, Chicago\, Illinois\, United States\, 60607 LOCATION:Room: Lecture Center C1\, Bldg: Lecture Center C1\, UIC east campu s\, Chicago\, Illinois\, United States\, 60607 ORGANIZER:devroye@uic.edu SEQUENCE:21 SUMMARY:IEEE Information Theory Society Distinguished Lecture URL;VALUE=URI:https://events.vtools.ieee.org/m/576026 X-ALT-DESC:Description: <br /><p>IEEE Information Theory Society Distinguis hed Lecture</p>\n<p>Title: Generative Modeling in Probability Space: From Wasserstein Optimization to Local Flow Matching</p>\n<p>Speaker:Yao Xie (< a href="https://www.isye.gatech.edu/users/yao-xie">Official webpage</a>) ( <a href="https://www2.isye.gatech.edu/~yxie77/YaoXie_CV.pdf">CV</a>) (<a h ref="https://www2.isye.gatech.edu/~yxie77/YaoXie_2page.pdf">2-page CV</a>) <br><a href="https://www.isye.gatech.edu/news/yao-xie-driving-data-science -innovation-coca-cola-foundation-chair">Coca-Cola Foundation Chair</a>&nbs p\;and Professor<br><a href="http://www.isye.gatech.edu/">H. Milton Stewar t School of Industrial and Systems Engineering (ISyE)</a><br>Associate Dir ector\,&nbsp\;<a href="https://ml.gatech.edu/">Machine Learning Center</a> <br>Adjunct Professor\,&nbsp\;<a href="https://www.ece.gatech.edu/">School of Electrical and Computer Engineering (ECE)</a><br><a href="http://www.g atech.edu/">Georgia Institute of Technology</a></p>\n<p>Abstract: How can we learn complex probability distributions through a sequence of simpler t ransformations with theoretical guarantees? This talk presents flow-based generative modeling as iterative computation in probability space\, connec ting optimization\, transport\, and diffusion. I will first discuss JKO fl ow models\, which implement proximal optimization in Wasserstein space. I will explain how convergence of the forward data-to-noise process yields g uarantees for reverse generation\, with bounds in Kullback-Leibler diverge nce that account for per-step optimization errors. I will then introduce L ocal Flow Matching\, which learns short transformations between nearby dis tributions along a diffusion path. This construction enables smaller model s and efficient training\, with generation guarantees in chi-square diverg ence derived from diffusion contraction. Together\, these results illustra te how optimization in probability space can guide the design and analysis of generative models. The talk will emphasize mathematical intuition and selected theoretical results\, complemented by numerical examples. This is based on joint work with Xiuyuan Cheng\, Jianfeng Lu\, Yixin Tan\, and Ch en Xu.</p>\n<p>Date: September 18\, 2026\, from 11:00am - 12:15pm Chicago time</p>\n<p>Location: UIC east campus\, lecture center C1\, Chicago\, IL 60607</p> END:VEVENT END:VCALENDAR