Abstract: To design fast neural networks, many works have been focusing on reducing the number of floating-point operations (FLOPs). We observe that such reduction in FLOPs, however, does not ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results