Speeding-up Continual Learning through Information Gains in Novel Experiences

Pierrick Lorang and Shivam Goel and Patrik Zips and Jivko Sinapov and Matthias Scheutz

We propose an integrated planning and learning approach that utilizes learning from failures and transferring knowledge over time to overcome novelty scenarios. The approach is more sample efficient in adapting to sudden and unknown changes (i.e., novelties) than the existing hybrid approaches. We showcase our results on a Minecraft-inspired gridworld environment called NovelGridworlds by injecting three novelties in the agent’s environment at test time. We show that our approach can speed up continual learning through information gained in each novel experience and, thus, more sample-efficient.

@inproceedings{lorangetal22prl,
  title={Speeding-up Continual Learning through Information Gains in Novel Experiences},
  author={Pierrick Lorang and Shivam Goel and Patrik Zips and Jivko Sinapov and Matthias Scheutz},
  year={2022},
  booktitle={Proceedings of 5th Workshop on Bridging the Gap Between AI Planning and Reinforcement Learning at International Joint Conference of Artificial Intelligence},
  url={https://hrilab.tufts.edu/publications/lorangetal22prl.pdf}
}