• Julien Martel

Our paper accepted @ICML2018

Updated: Jun 26, 2018

Our paper with Lorenz and Giacomo got accepted at ICML 2018 and will be presented in Stockholm in July!

The paper presents a method to reparametrize weight matrices of an artificial neural network by a kernel function. Each weight in the matrix is the result of a kernel application between two low-dimensional vectors of parameters that can be thought as some location/centroids in a low-dimensional embedding space. We show how this can be beneficial for data-visualization, to constrain a matrix to be low-rank and present state-of-the-art results on the Movie-Lens Dataset (a recommendation task in which ratings of movies need to be predicted for unknown users).


2019 © Julien N. P. MARTEL | Stanford, USA | jnmartel [at] stanford [dot] edu