Learning sparse and meaningful representations through embodiment
Standard
Learning sparse and meaningful representations through embodiment. / Clay, Viviane; König, Peter; Kühnberger, Kai-Uwe; Pipa, Gordon.
in: NEURAL NETWORKS, Jahrgang 134, 02.2021, S. 23-41.Publikationen: SCORING: Beitrag in Fachzeitschrift/Zeitung › SCORING: Zeitschriftenaufsatz › Forschung › Begutachtung
Harvard
APA
Vancouver
Bibtex
}
RIS
TY - JOUR
T1 - Learning sparse and meaningful representations through embodiment
AU - Clay, Viviane
AU - König, Peter
AU - Kühnberger, Kai-Uwe
AU - Pipa, Gordon
N1 - Copyright © 2020 The Authors. Published by Elsevier Ltd.. All rights reserved.
PY - 2021/2
Y1 - 2021/2
N2 - How do humans acquire a meaningful understanding of the world with little to no supervision or semantic labels provided by the environment? Here we investigate embodiment with a closed loop between action and perception as one key component in this process. We take a close look at the representations learned by a deep reinforcement learning agent that is trained with high-dimensional visual observations collected in a 3D environment with very sparse rewards. We show that this agent learns stable representations of meaningful concepts such as doors without receiving any semantic labels. Our results show that the agent learns to represent the action relevant information, extracted from a simulated camera stream, in a wide variety of sparse activation patterns. The quality of the representations learned shows the strength of embodied learning and its advantages over fully supervised approaches.
AB - How do humans acquire a meaningful understanding of the world with little to no supervision or semantic labels provided by the environment? Here we investigate embodiment with a closed loop between action and perception as one key component in this process. We take a close look at the representations learned by a deep reinforcement learning agent that is trained with high-dimensional visual observations collected in a 3D environment with very sparse rewards. We show that this agent learns stable representations of meaningful concepts such as doors without receiving any semantic labels. Our results show that the agent learns to represent the action relevant information, extracted from a simulated camera stream, in a wide variety of sparse activation patterns. The quality of the representations learned shows the strength of embodied learning and its advantages over fully supervised approaches.
U2 - 10.1016/j.neunet.2020.11.004
DO - 10.1016/j.neunet.2020.11.004
M3 - SCORING: Journal article
C2 - 33279863
VL - 134
SP - 23
EP - 41
JO - NEURAL NETWORKS
JF - NEURAL NETWORKS
SN - 0893-6080
ER -