Not really, no. Especially not if this is implemented in a specialized accelerator. A GFLOP is not that much there. Also, like most other neural network algorit

Not really, no. Especially not if this is implemented in a specialized accelerator. A GFLOP is not that much there. Also, like most other neural network algorithms, this could also be done in fixed point, thereby further reducing the computational cost.