In recent years, research has shown that transfer learning methods can be leveraged to construct curricula that sequence a series of simpler tasks such that performance on a final target task is improved. A major limitation of existing approaches is that such curricula are handcrafted by humans that are typically domain experts. To address this limitation, we introduce a method to generate a curriculum based on task descriptors and a novel metric of transfer potential. Our method automatically generates a curriculum as a directed acyclic graph (as opposed to a linear sequence as done in existing work). Experiments in both discrete and continuous domains show that our method produces curricula that improve the agent's learning performance when compared to the baseline condition of learning on the target task from scratch.

PDF, HTML

In Proceedings of the 31st AAAI Conference on Artificial Intelligence (AAAI), San Francisco, CA, February 2017.

@inproceedings{AAAI17-Svetlik, title={Automatic Curriculum Graph Generation for Reinforcement Learning Agents}, author={Maxwell Svetlik and Matteo Leonetti and Jivko Sinapov and Rishi Shah and Nick Walker and Peter Stone}, booktitle={Proceedings of the 31st AAAI Conference on Artificial Intelligence (AAAI)}, month={February}, address={San Francisco, CA}, url="http://nn.cs.utexas.edu/?svetlik:aaai17", year={2017} }