Divide (or Convolve) and Conquer: Data Comparison with Wiener Filters

Explore a novel approach to data comparison using Wiener filters in this 51-minute talk by Lluis Guasch. Delve into the limitations of current methods for comparing data samples and discover how Wiener-filter theory can address these challenges. Learn about the convolutional nature of Wiener filters and their ability to globally compare elements between samples. Examine the validation of this approach through four machine learning applications: data compression, medical imaging imputation, translated classification, and non-parametric generative modeling. Gain insights into how this method achieves increased resolution in reconstructed images with improved perceptual quality and demonstrates robustness against translations compared to conventional mean-squared-error implementations.