BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Causal Reasoning from Meta-reinforcement Learning - Jakub Perlin (
 University of Cambridge)
DTSTART:20190313T170000Z
DTEND:20190313T190000Z
UID:TALK121096@talks.cam.ac.uk
CONTACT:Adrià Garriga Alonso
DESCRIPTION:This week we read “Causal Reasoning from Meta-reinforcement 
 Learning” by Ishita Dasgupta et al. (https://arxiv.org/abs/1901.08162). 
 Please read the paper before coming.\n\nAs usual\, there will be free pizz
 a. The first half hour is for stragglers to finish reading.\n\n*Abstract:*
 \n\n     Discovering and exploiting the causal structure in the environmen
 t is a crucial challenge for intelligent agents. Here we explore whether c
 ausal reasoning can emerge via meta-reinforcement learning. We train a rec
 urrent network with model-free reinforcement learning to solve a range of 
 problems that each contain causal structure. We find that the trained agen
 t can perform causal reasoning in novel situations in order to obtain rewa
 rds. The agent can select informative interventions\, draw causal inferenc
 es from observational data\, and make counterfactual predictions. Although
  established formal causal reasoning algorithms also exist\, in this paper
  we show that such reasoning can arise from model-free reinforcement learn
 ing\, and suggest that causal reasoning in complex settings may benefit fr
 om the more end-to-end learning-based approaches presented here. This work
  also offers new strategies for structured exploration in reinforcement le
 arning\, by providing agents with the ability to perform -- and interpret 
 -- experiments. \n\nInvite your friends to join the mailing list (https://
 lists.cam.ac.uk/mailman/listinfo/eng-safe-ai)\, the Facebook group (https:
 //www.facebook.com/groups/1070763633063871) or the talks.cam page (https:/
 /talks.cam.ac.uk/show/index/80932). Details about the next meeting\, the w
 eek’s topic and other events will be advertised in these places.\n
LOCATION:Engineering Department\, CBL Seminar room BE4-38
END:VEVENT
END:VCALENDAR