About
Robots and other artificial agents can learn to solve tasks with superhuman performance, however when faced with a different task, the learning process starts again from scratch. Benjamin Rosman’s research asks how the knowledge gained from solving one task can be reused in the next, by drawing insight from humans’ ability to decompose complex problems into reusable pieces. So, if a robot has previously mastered the skills of opening doors and carrying coffee, it should be able to reuse these to solve later problems, including doing them simultaneously. Using this decomposition also allows humans to safely specify required goals, and easily interpret the robot’s behaviour.
Awards
- Faculty of Science Supervisor Award, University of the Witwatersrand, 2022
- Friedel Sellschop Award, University of the Witwatersrand, 2022
- Faculty Research Award, Google, 2018
- Young Researchers Establishment Fund, CSIR, 2014
Relevant Publications
- Nangue Tasse, G., James, S., & Rosman, B. (2020). A Boolean task algebra for reinforcement learning. Advances in Neural Information Processing Systems, 33, 9497-9507.
- Van Niekerk, B., James, S., Earle, A., & Rosman, B. (2019, May). Composing value functions in reinforcement learning. In International Conference on Machine Learning (pp. 6401-6409). PMLR.
- Rosman, B., Hawasly, M., & Ramamoorthy, S. (2016). Bayesian policy reuse. Machine Learning, 104(1), 99-127.