BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Symmetries in Reinforcement Learning - Robert Pinsler and Adria Ga
 rriga Alonso (University of Cambridge)
DTSTART:20201118T110000Z
DTEND:20201118T123000Z
UID:TALK153898@talks.cam.ac.uk
CONTACT:Elre Oldewage
DESCRIPTION:In recent years\, there has been great interest in learning an
 d\nexploiting symmetries of various machine learning problems. For example
 \,\nthis led to a generalization of CNNs originally designed for\ntranslat
 ion-invariant image classification in Euclidean space to much\nmore genera
 l manifolds and graphs. The goal of this talk is to give an\noverview of a
 pproaches that specifically deal with symmetries in\nreinforcement learnin
 g problems\, e.g. as commonly encountered in\ncontrol\, medicine or chemis
 try. In the first part of this talk\, we will\nuse group theory to charact
 erize and formalize such symmetries using the\nnotion of MDP homomorphisms
 . In the second part\, we will present recent\ndeep reinforcement learning
  methods for learning and exploiting MDP\nhomomorphisms.\n\nRecommended re
 ading:\n1) van der Pol\, Elise\, et al. "Plannable Approximations to MDP\n
 Homomorphisms: Equivariance under Actions." arXiv preprint\narXiv:2002.119
 63 (2020).\n2) van der Pol\, Elise\, et al. "MDP homomorphic networks: Gro
 up\nsymmetries in reinforcement learning." Advances in Neural Information\
 nProcessing Systems 33 (2020).
LOCATION:https://eng-cam.zoom.us/j/86068703738?pwd=YnFleXFQOE1qR1h6Vmtwbno
 0LzFHdz09
END:VEVENT
END:VCALENDAR