MNIST Classification
The simulated three-layer classifier reaches 91.2% test accuracy on MNIST. The reported accuracy remains substantially below near-perfect electronic neural-network performance on MNIST. The MNIST result is interpreted as showing that passi…
1 sources - 5 claims
The simulated three-layer classifier reaches 91.2% test accuracy on MNIST. The reported accuracy remains substantially below near-perfect electronic neural-network performance on MNIST. The MNIST result is interpreted as showing that passive diffractive propagation can separate handwritten digit class distributions despite its restricted computational form. Scaling beyond MNIST is unresolved and may require larger or richer optical architectures. The main experimental setup uses MNIST as a controlled benchmark for the diffractive stack.