Data-efficient Modeling of Optical Matrix Multipliers Using Transfer Learning
This work addresses data scarcity in modeling optical computing devices, offering a more efficient approach for researchers and engineers in photonics and optical computing, though it appears incremental as it builds on existing transfer learning methods.
The authors tackled the problem of modeling optical matrix multipliers with limited experimental data by using transfer learning-assisted neural networks, achieving performance that surpasses analytical models while requiring less than 10% of the data needed for optimal results.
We demonstrate transfer learning-assisted neural network models for optical matrix multipliers with scarce measurement data. Our approach uses <10\% of experimental data needed for best performance and outperforms analytical models for a Mach-Zehnder interferometer mesh.