Convolutional neural networks for sequential data and tropical quasisymmetric functions

The success of CNNs in image recognition is attributed to weight sharing and their structural compatibility with image data (approximate translation invariance, perceptive field, ..). Aiming for a structure that is relevant to sequential data (and is sharing weights), we lay the mathematical groundwork by extending the notion of quasisymmetric functions to semirings. All objects will be introduced and the talk will be self-contained. I briefly outline how this can fit into a deep learning pipeline.

This is joint work with Kurusch Ebrahimi-Fard (NTNU Trondheim) and Nikolas Tapia (WIAS Berlin).