Double Deep Q Network with In-Parallel Experience Generator

In this paper, an algorithm, for in-parallel, greedy experience generator (briefly IPE, In Parallel Experiences), has been crafted, and added to the Double Deep Q-Learning algorithm. The algorithm aims to perturbs the weights of the online network, and as results, the network, trying to recover from the perturbed weights, escapes from the local minima. DDQN with IPE takes about the double of time of the previous to compute, but even if it slows down the learning rate in terms of wall clock time, the solution converges faster in terms of number of epochs.