Hardware-In-The-Loop Training of a 4f Optical Correlator with Logarithmic Complexity Reduction for CNNs

This work evaluates a forward-only learning algorithm on the MNIST dataset with hardware-in-the-loop training of a 4f optical correlator, achieving 87.6% accuracy with $O\left(n^{2}\right)$ complexity, compared to backpropagation, which achieves 88.8% accuracy with $O\left(n^{2} \log n\right)$ complexity.