CST-RL: Contrastive Spatio-Temporal Representations for Reinforcement Learning

• Published in: IEEE access Vol. 11; p. 1
• Main Authors: Ho, Chi-Kai, King, Chung-Ta
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.01.2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: 3D CNNs; Artificial neural networks; contrastive learning; Control tasks; Correlation; Feature extraction; Machine learning; Pixels; Reinforcement learning; Representation learning; Representations; sample efficiency; Spatial data; spatio-temporal representation learning; Spatiotemporal phenomena; Standard scores; Task analysis; Three-dimensional displays; Training

Learning representations from high-dimensional observations is critical for training of pixel-based continuous control tasks with reinforcement learning (RL). Without proper representations, the training will be very inefficient, requiring long training time and huge training data to learn directly...