BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Kinetic energy choice in Hamiltonian/hybrid Monte Carlo - Sam Livi
 ngstone (University of Bristol)
DTSTART:20170705T100000Z
DTEND:20170705T104500Z
UID:TALK73154@talks.cam.ac.uk
CONTACT:INI IT
DESCRIPTION:<span>We consider how different choices of kinetic energy in H
 amiltonian Monte Carlo affect algorithm performance. To this end\, we intr
 oduce two quantities which can be easily evaluated\, <i>the composite grad
 ient</i> and<i> the implicit noise</i>. Results are established on integra
 tor stability and geometric convergence\, and we show that choices of kine
 tic energy that result in heavy-tailed momentum distributions can exhibit 
 an undesirable <i>negligible moves</i> property\, which we define. A gener
 al efficiency-robustness trade off is outlined\, and implementations which
  rely on approximate gradients are also discussed. Two numerical studies i
 llustrate our theoretical findings\, showing that the standard choice whic
 h results in a Gaussian momentum distribution is not always optimal in ter
 ms of either robustness or efficiency.</span>
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR