Functional Deep Neural Network

Functional deep neural networks (FDNNs) are a class of neural networks designed to process entire functions or other neural networks as input, rather than individual data points. Current research focuses on developing efficient and expressive FDNN architectures, often incorporating principles of equivariance to symmetries within the input network's weights, and employing techniques like kernel embedding for dimensionality reduction. These models find applications in diverse fields, including implicit neural representation processing, learned optimization, and classification tasks involving complex data such as EEG signals and microscopy images, offering powerful tools for analyzing and interpreting high-dimensional functional data. The development of robust and efficient FDNNs promises significant advancements in various scientific domains and practical applications.