BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20191206T142454Z
LOCATION:Analog 1\, 2
DTSTART;TZID=Europe/Stockholm:20180626T113000
DTEND;TZID=Europe/Stockholm:20180626T120000
UID:isc_hpc_ISC High Performance 2018_sess110_pap133@linklings.com
SUMMARY:Distributed Deep Reinforcement Learning: Learn how to Play Atari G
 ames in 21 Minutes
DESCRIPTION:Research Paper\n\nDistributed Deep Reinforcement Learning: Lea
 rn how to Play Atari Games in 21 Minutes\n\nGrel, Michalewski, Kaczmarek, 
 Jędrych, Adamski...\n\nWe present a study in Distributed Deep Reinfor
 cement Learning (DDRL) focused on scalability of a state-of-the-art Deep R
 einforcement Learning algorithm known as Batch Asynchronous Advantage Acto
 r-Critic (BA3C). We show that using the Adam optimization algorithm with a
  batch size of up to 2048 is a viable choice for carrying out large scale 
 machine learning computations. This, combined with careful reexamination o
 f the optimizer’s hyperparameters, using synchronous training on the node 
 level (while keeping the local, single node part of the algorithm asynchro
 nous) and minimizing the memory footprint of the model, allowed us to achi
 eve linear scaling for up to 64 CPU nodes. This corresponds to a training 
 time of 21 minutes on 768 CPU cores, as opposed to 10 hours when using a s
 ingle node with 24 cores achieved by a baseline single-node implementation
 .\n\nPasses: Conference Pass, AI/Machine Learning/Deep Learning, Clouds an
 d Distributed Computing, Communication Optimization, Parallel Algorithms\n
 \nTag: Conference Pass, AI/Machine Learning/Deep Learning, Clouds and Dist
 ributed Computing, Communication Optimization, Parallel Algorithms
URL:https://2018.isc-program.com/presentation/?id=pap133&sess=sess110
END:VEVENT
END:VCALENDAR