Fuzzy Deep Reinforcement Learning

Inspired by the recent variants in Deep RL, Fuzzy Deep RL is developed by using Fuzzy Logic as a data representation method and Deep Q-Networks for autoscaling problem

References:

The used dataset from Clarknet Traces ftp://ita.ee.lbl.gov/traces/

Simulation code refined from Feedback Control for Computer Systems by Philipp K. Janert (O'Reilly Media)

Fuzzy Q-Learning corrected from FQL

Deep QN architecture derived from Hands-on Deep Reinforcement Learning, published by Packt

For the use of fuzzy logic with DQN, please see Fuzzy DQN and Fuzzy Q Learning FQL

If you use this code please cite as:

Doan D.N., Zaharie D., Petcu D. (2020) Auto-scaling for a Streaming Architecture with Fuzzy Deep Reinforcement Learning. In: Schwardmann U. et al. (eds) Euro-Par 2019: Parallel Processing Workshops. Euro-Par 2019. Lecture Notes in Computer Science, vol 11997. Springer, Cham. https://doi.org/10.1007/978-3-030-48340-1_37