Adapting to Unseen Environments through Explicit Representation of Context by Tutum, C., & Miikkulainen, R. (2020)

source

(Tutum and Miikkulainen 2020)

tags

Meta-learning, Reinforcement learning, ALife 2020

Summary

This work introduces the idea of Context-Skill networks for continuous RL tasks. Experiments are done on a Flappy bird like game.

The authors use a LSTM as a context network to make part of the prediction and a feed-forward neural network as a skill network. They are able to demonstrate that in that game, better performances are achieved by using both networks compared to a single one.

Bibliography

1. Tutum, Cem, and Risto Miikkulainen. July 1, 2020. "Adapting to Unseen Environments through Explicit Representation of Context". Artificial Life Conference Proceedings 32 (July). MIT Press:581–88. DOI.