Fitness a novelty v evolučním zpětnovazebném učení
| Thesis title in Czech: | Fitness a novelty v evolučním zpětnovazebném učení |
|---|---|
| Thesis title in English: | Fitness and novelty in evolutionary reinforcement learning |
| Key words: | Evoluční algoritmy|zpětnovazebné učení|hledání novelty|strategie explorace |
| English key words: | Evolutionary algorithms|reinforcement learning|novelty search|exploration strategies |
| Academic year of topic announcement: | 2023/2024 |
| Thesis type: | diploma thesis |
| Thesis language: | |
| Department: | Department of Theoretical Computer Science and Mathematical Logic (32-KTIML) |
| Supervisor: | Mgr. Roman Neruda, CSc. |
| Author: | hidden - assigned and confirmed by the Study Dept. |
| Date of registration: | 05.12.2023 |
| Date of assignment: | 07.12.2023 |
| Confirmed by Study dept. on: | 07.12.2023 |
| Guidelines |
| Evolutionary optimization techniques are one of the effective methods for solving feedback learning problems. Recently, the use of behavioral characteristics and novelty has been studied to strengthen exploration and avoid getting stuck in the local extremes of the objective function.
The aim of the thesis is to investigate different strategies for novelty search within evolutionary algorithms applied to reinforcement learning problems. The student will implement several evolutionary algorithms suitable for continuous optimizations (evolutionary strategy, differential evolution). As part of his own work, he will propose a novel representation of agents' behavior in reinforcement learning. One of the problems studied will be the mutual influence of fitness and novelty during agent learning. The student will experimentally verify the effectiveness of the proposed algorithms on several typical feedback learning tasks (Cart pole, Lunar lander, etc.). |
| References |
| Ahmed Hallawa, Thorsten Born, Anke Schmeink, Guido Dartmann, Arne Peine, Lukas Martin, Giovanni Iacca, A. Eiben, Gerd Ascheid. (2021). Evo-RL: evolutionary-driven reinforcement learning. GECCO '21: Genetic and Evolutionary Computation Conference, Companion Volume, 153-154. 10.1145/3449726.3459475.
Joel Lehman and Kenneth O. Stanley. 2011. Abandoning Objectives: Evolution through the Search for Novelty Alone. Evolutionary Computation journal, (19):2, pages 189-223, Cambridge, MA: MIT Press. Elliot Meyerson, Joel Lehman, and Risto Miikkulainen. 2016. Learning Behavior Characterizations for Novelty Search. In Proceedings of the Genetic and Evolutionary Computation Conference 2016 (GECCO '16). Association for Computing Machinery, New York, NY, USA, 149–156. https://doi.org/10.1145/2908812.2908929 Ethan C. Jackson and Mark Daley. 2019. Novelty search for deep reinforcement learning policy network weights by action sequence edit metric distance. In Proceedings of the Genetic and Evolutionary Computation Conference Companion (GECCO '19). Association for Computing Machinery, New York, NY, USA, 173–174. https://doi.org/10.1145/3319619.3321956 Stephane Doncieux and Giuseppe Paolo and Alban Laflaquière and Alexandre Coninx. 2020. Novelty Search makes Evolvability Inevitable. arXiv, 2005.06224, 2020. https://arxiv.org/abs/2005.06224 |