I am using a NN within a reinforcement learning context where the NN estimates the expected reward for a given action/enviroment touple. For this reason i need a single prediction at a time. My understanding from parallelisation in Deep Learning comtext is, that usually the parallelisation is between predictions which is in my context unforunately not possible.
Is it possible to use parallelisation even for single predictions (through Independent parallel nodes in the NN).