An Investigation into Pre-Training Object-Centric Representations for Reinforcement Learning.
Jaesik YoonYi-Fu WuHeechul BaeSungjin AhnPublished in: ICML (2023)
Keyphrases
- reinforcement learning
- moving objects
- object representation
- training set
- d objects
- object description
- supervised learning
- training phase
- object model
- test set
- geometric models
- object models
- training process
- function approximation
- markov decision processes
- neural network
- training examples
- state space
- image features
- data objects
- spatial relations
- spatial relationships
- higher level
- object segmentation
- optimal control
- training samples
- multiple objects
- target object
- optimal policy
- support vector
- multi agent
- computer vision
- function approximators
- learning algorithm
- machine learning