Table 1 Comparison of the numerical performances of different optimization frameworks
From: Non-convex optimization for inverse problem solving in computer-generated holography
| Â | Reconstruction performance | Time | Memory | Loss function | Gradient | |
|---|---|---|---|---|---|---|
Alternating Projection | FFT | Moderate accuracy | Fast | Small | Not required | None |
ASM | Distance-dependent | Fast | Moderate | |||
SGD | FFT | Unstable convergence | Very fast | Moderate | Required | First-order |
ASM | Distance-dependent | Fast | Moderate | |||
Quasi-Newton | FFT | High accuracy | Slow | Large | Required | Second-order |
ASM | Distance-dependent | Slow | Large | |||
