How does Reinforcement Learning (RL) relate to Metaheuristic Search (MS) techniques like Genetic Algorithm, Particle Swarm Optimization, etc. and to Generative Adversarial Learning used in GANs? I feel that MS is similar to RL in the sense that it creates the next population based on feedback about the suitability of candidates from the present population based on function evaluations. Similarly, the generator and discriminator networks in GANs train based on feedback from the each other. What are your views? What are some other methods apart from MS and GANs that are similar to RL but not consider to be RL exactly?
Dear Shounak,
It is a very interesting question. The Q-learning algorithm can be represented as an optimization problem (maximization problem), where the candidate solutions are the Q-values and the fitness function is the Q-function.
For more details about this topic:
Article Action-Selection Method for Reinforcement Learning Based on ...
Article Bat Q-learning Algorithm
Dear Bilal,
Thanks for the informative answer. I think Q-learning is rather interesting and will have a look at the articles suggested by you.
- Badges
- Science topic