MEAHO: Membrane Evolutionary Algorithm for Hyperparameter Optimization of Deep Convolutional Neural Networks

Membrane algorithm has been used to solve many optimization problems since it was put forward. These methods used the membrane algorithm as a container for other algorithms to solve many problems, such as, traveling salesman problem, the knapsack problem and so on. In this paper, from the angle of membrane algorithm, the solution of hyperparameter optimization problem by membrane algorithm is explored, with the hyperparameter space divided by lattice method, which transformed into membrane structure subsequently. For the evolution of membrane, the entropy reduction principle is proposed. Compared with other membrane algorithms, this paper solves the optimization problem from the membrane algorithm itself, designs the corresponding algorithm for the membrane structure, and obtains the experimental results through CNN experiment.