Answers are grounded in this paper and cite their sources.
We use octave-spanning, phase-shaped supercontinuum generation as an analog computing element in a neural network. We can perform standard machine learning tasks such as autoencoding by embedding the physical device within a virtual neural network which converts input data to optical parameters, and measured spectra to output data. To understand the computing potential of the supercontinuum, we fully measure a small subset of input parameter space. We test universal function approximation by forming a basis set of delta functions from linear combinations of the physically measured functions. This approach provides a more general way to estimate the computational power of a physical operation than performing a specific task.
No takes yet. Share an insight, caveat, or question.
Lee et al. (2024) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: