Table 6:
Analysis of the computational cost between CAML and MultiResCNN. “m”, “s”, “ep” and “d” denote million, second, epoch and document respectively.
| CAML | MultiResCNN | |
|---|---|---|
| Parameter Amount | 6.2m | 11.9m |
| Training Time | 438s/ep | 1026s/ep |
| Training Epoch | 85 | 26 |
| Inference Speed | 108.7d/s | 70.9d/s |