Low-Precision Reinforcement Learning:
Running Soft Actor-Critic in Half Precision
Johan Bjorck 1 Xiangyu Chen 1 Christopher De Sa 1 Carla P. Gomes 1 Kilian Q. Weinberger 1
Abstract learning, an emerging trend for accelerating deep learning
Low-precision training has become a popular ap- is low-precision training, where one uses fewer than 32
proach to reduce compute requirements, memory bits to represent individual parameters, ...
附件列表