BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Approaches to avoiding negative side effects - Adrià Garriga Alon
 so (University of Cambridge)
DTSTART:20180530T160000Z
DTEND:20180530T173000Z
UID:TALK106537@talks.cam.ac.uk
CONTACT:Adrià Garriga Alonso
DESCRIPTION:In this session we will learn about several approaches to avoi
 ding negative side-effects\, from the papers:\n* "Minimax-Regret *Querying
  on Side Effects* for Safe Optimality in Factored Markov Decision Processe
 s"\, Zhang et al. 2018 (emphasis mine)\n* "Low Impact Artificial Intellige
 nces"\, Armstrong and Levinstein 2017\n\nThe first paper's approach is rea
 sonably efficient to compute. However\, it only applies to discrete-state 
 factored MDPs\, the human feedback it requires probably doesn't scale grea
 t\, and it doesn't account for all kinds of positive or negative side effe
 cts.\n\nThe approaches from the second paper are less immediately applicab
 le and difficult to compute. Both provide some insights\, and we will base
  our discussion of how to improve side-effect measures on them.\n\nRelevan
 t papers:\n\n"Minimax-Regret Querying on Side Effects for Safe Optimality 
 in Factored Markov Decision Processes"\, Shun Zhang\, Edmund H. Durfee\, a
 nd Satinder Singh\, 2018\, https://web.eecs.umich.edu/~baveja/Papers/ijcai
 -2018.pdf\n\n"Low Impact Artificial Intelligences"\, Armstrong and Levinst
 ein 2017 https://arxiv.org/abs/1705.10720\n\n"AI Safety Gridworlds"\, Leik
 e et al. 2017\, https://arxiv.org/abs/1711.09883\n\n"Concrete Problems in 
 AI Safety"\, Amodei et al. 2016 https://arxiv.org/abs/1606.06565
LOCATION: Cambridge University Engineering Department\, CBL Seminar room B
 E4-38.  For directions see http://learning.eng.cam.ac.uk/Public/Directions
END:VEVENT
END:VCALENDAR