Solving 100-Digit Challenge using Q-learning inspired Differential Evolution Algorithm
Prathu Bajpai, Jagdish Chand Bansal · 2024
The 100-digit challenge requires solving ten complex optimization problems of the CEC 2019 benchmark test suite with the precision of ten digits. The challenge does not recommend any constraint on the maximum number of function evaluations or iterations. The rationale behind this relaxation is to reward accuracy over rapid convergence. This research proposes a Q-learning inspired Differential Evolution (DE) (QLiDE) algorithm to solve the 100-digit challenge. A Q-learning framework consisting of state-space and action-space is set up using the fitness vector and a pool of DE mutation strategies, respectively. The mutation strategies in the pool are rewarded based on their cumulative performance, and the reward values are utilized as feedback signals for updating the Q-values. The proposed algorithm achieves the score of 93 out of 100 and secures an overall rank 3 over all other algorithms applied for solving the 100-digit challenge, till date.