toward object-oriented deep reinforcement...
TRANSCRIPT
Matthew BotvinickDeepMind, London UKGatsby Computational Neuroscience Unit, UCL
Toward object-oriented deep reinforcement learning
+1
+1
atari
Mnih et al, Nature (2015)
+1
+1
Jaderberg et al., Science, 2019
+1
+1
dqn convnet
Mnih et al, Nature (2015)
+1
+1lake
+1
+1
+1
+1objects — pic
+1
+1human objects
Kahneman et al., 1992
+1
+1
Egly, Driver, and Rafal (1994); Moore, Yantis, and Vaughan (1998)
Automatic spread of attention
+1
+1
Roelfsema et al. Nature, 1998
+1
+1
LO??? (Kanwisher)
Malach et al., PNAS, 1995
+1
+1
objects — pic AGAIN
+1
+1diuk (cf?)
cf. Keramati et al., 2018; Cobo et al., 2013; Garnelo et al., 2016; Lazaro-Gradillo et al., 2019; Zambaldi, et al., 2018
+1
+1
+1
+1
E.g., Girshick, 2015; He et al., 2017; Redmon & Farhadi, 2018
+1
+1
Alex Lerchner Chris Burgess Loic Matthey Klaus Greff
Nick Watters Irina Higgins Rishabh Kabra Malcolm Reynolds
half refrigerator
other half refrigerator
+1
+1
objects — pic AGAIN
+1
+1
+1
+1
+1
+1
+1
+1
+1
Kahneman & Treisman, 1984: Object Files
Green, Edwin James, and Jake Quilty-Dunn. "what is an object file?." The British Journal for the Philosophy of Science (2017).
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
Alex Lerchner Chris Burgess Loic Matthey Klaus Greff
Nick Watters Irina Higgins Rishabh Kabra Malcolm Reynolds