Moritz Lange (Institute for Neural Computation, Ruhr-University Bochum) and Raphael Engelhardt (Cologne Institute of Computer Science, TH Köln) won the Best Paper Award at the 9th International Conference on Machine Learning, Optimization, and Data Science (LOD 2023) for their work “Improving Reinforcement Learning Efficiency with Auxiliary Tasks in Non-Visual Environments: A Comparison”. Their research carries significant implications for practical reinforcement learning (RL) applications in various industries.
Notable Findings:
- The study is the first to empirically compare various conventional representation learning approaches for reinforcement learning in a systematical way.
- Their research underscores the importance of learning environment dynamics as a key factor for enhancing RL performance. This is obtained by introducing an auxiliary task to the original task of optimizing RL rewards.
These groundbreaking insights open the door to more efficient and practical RL solutions, particularly in environments where visual observations are limited.
Congratulations to Moritz and Raphael!
Full reference:
Lange, M., Krystiniak, N., Engelhardt, R. C., Konen, W., & Wiskott, L. (2023). Improving Reinforcement Learning Efficiency with Auxiliary Tasks in Non-Visual Environments: A Comparison.
Click to show paper on arXiv