BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
PRODID:iCalendar-Ruby
BEGIN:VEVENT
CATEGORIES:
DESCRIPTION:As the continuous limit of many discretized algorithms\, PDEs c
an provide a qualitative description of algorithm's behavior and give princ
ipled theoretical insight into many mysteries in machine learning. In this
talk\, I will give a theoretical interpretation of several machine learning
algorithms using Fokker-Planck (FP) equations. In the first one\, we provi
de a mathematically rigorous explanation of why resampling outperforms rewe
ighting in correcting biased data when stochastic gradient-type algorithms
are used in training. In the second one\, we propose a new method to allevi
ate the double sampling problem in model-free reinforcement learning\, wher
e the FP equation is used to do error analysis for the algorithm. In the la
st one\, inspired by an interactive particle system whose mean-field limit
is a non-linear FP equation\, we develop an efficient gradient-free method
that finds the global minimum exponentially fast.
DTEND:20220222T151500Z
DTSTAMP:20241016T053239Z
DTSTART:20220222T140000Z
LOCATION:Maxwell Dworkin G115
SEQUENCE:0
SUMMARY:Fokker-Planck Equations and Machine Learning
UID:tag:localist.com\,2008:EventInstance_39127089319917
URL:https://calendar.college.harvard.edu/event/fokker-planck_equations_and_
machine_learning
END:VEVENT
END:VCALENDAR