Measure gradients, not activations! Enhancing neuronal activity in deep reinforcement learning | Read Paper on Bytez