BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:The master equation for the reinforcement learning
DTSTART;VALUE=DATE-TIME:20190226T173600Z
DTEND;VALUE=DATE-TIME:20190226T174800Z
DTSTAMP;VALUE=DATE-TIME:20190616T163342Z
UID:indico-contribution-154@c-tau.ru
DESCRIPTION:Speakers: Edgar Vardanyan (Yerevan Physics Institute\, Yerevan
State University)\nWe look the reinforcement learning dynamics. As the dy
namics is a stochastic process\, the adequate mathematical tool is the mas
ter equation. We introduce the probability distributions for the actions a
nd value functions\, then get a master equation\, describing the reinforce
ment learning process. We derived a Hamilton-Jacobi equation for the latte
r equation. We verify a unique feature of the model (compared to the Maste
r equation of the chemical reaction with few molecules or evolution models
with finite population): the variance of distribution disappeared at the
steady state\, which gives a good credit for the application of the moment
closing approximation. Our method (recursive equations) gives accurate ex
pressions both for the mean and variance of variables\, while HJE provides
only correct results for the mean values. Looking the recursive equations
\, we express the value function distribution via the solution of a system
of ordinary differential equations.\n\nhttps://mosphys.ru/indico/event/2/
contributions/154/
LOCATION:HSE Study Center “Voronovo”
URL:https://mosphys.ru/indico/event/2/contributions/154/
END:VEVENT
END:VCALENDAR