Relational Deep Reinforcement Learning


Relational information is important in some reinforcement learning tasks.


image Relational information can be important source for high scores in reinforcement learning. To provide inductive bias for relational information, this paper appied self-attention(MHDPA) to the last layer of the convolution network that encode the state. Another multilayer perceptron(f_\theta) is applied in parallel to each self-attended objects.


The agent achieved state-of-the-art result in Box-World and StarCrat2 Minigame.


Good inductive bias for relational information which is useful for agent.

Zambaldi, Vinicius, et al. “Relational Deep Reinforcement Learning.” arXiv preprint arXiv:1806.01830 (2018).

© 2017. by isme2n

Powered by aiden