Fig. 3
From: Neural network compression for reinforcement learning tasks
General scheme of training. A randomly initialized neural network is trained for 20% of the total steps in a classical manner. Further, during the 20-80% of training, gradual pruning with n steps is applied. Then pruning is turned off and from 80 to 100% of steps the network is trained again in the classical way. If quantization is required, an additional 20% of training steps (from 100% to 120%) are performed with 8-bit quantization.
