Publications by Guillermo Ortiz Jimenez

The inductive bias of deep learning: Connecting weights and functions

Years of a fierce competition have naturally selected the fittest deep learning algorithms. Yet, although these models work well in practice, we still lack a proper characterization of why they do so. This poses serious questions about the robustness, trus ...

EPFL2023

A Structured Dictionary Perspective on Implicit Neural Representations

Pascal Frossard, Guillermo Ortiz Jimenez, Beril Besbinar, Gizem Yüce

Implicit neural representations (INRs) have recently emerged as a promising alternative to classical discretized representations of signals. Nevertheless, despite their practical success, we still do not understand how INRs represent signals. We propose a ...

IEEE COMPUTER SOC2022

PRIME: A Few Primitives Can Boost Robustness to Common Corruptions

Pascal Frossard, Seyed Mohsen Moosavi Dezfooli, Guillermo Ortiz Jimenez, Apostolos Modas

Despite their impressive performance on image classification tasks, deep networks have a hard time generalizing to unforeseen corruptions of their data. To fix this vulnerability, prior works have built complex data augmentation strategies, combining multi ...

SPRINGER INTERNATIONAL PUBLISHING AG2022

What can linearized neural networks actually say about generalization?

Pascal Frossard, Seyed Mohsen Moosavi Dezfooli, Guillermo Ortiz Jimenez

For certain infinitely-wide neural networks, the neural tangent kernel (NTK) theory fully characterizes generalization, but for the networks used in practice, the empirical NTK only provides a rough first-order approximation. Still, a growing body of work ...

2021

Forward-Backward Splitting for Optimal Transport Based Problems

Pascal Frossard, Guillermo Ortiz Jimenez, Mireille El Gheche, Hermina Petric Maretic, Effrosyni Simou

Optimal transport aims to estimate a transportation plan that minimizes a displacement cost. This is realized by optimizing the scalar product between the sought plan and the given cost, over the space of doubly stochastic matrices. When the entropy regula ...

IEEE2020

Neural Anisotropy Directions

Pascal Frossard, Seyed Mohsen Moosavi Dezfooli, Guillermo Ortiz Jimenez, Apostolos Modas

In this work, we analyze the role of the network architecture in shaping the inductive bias of deep classifiers. To that end, we start by focusing on a very simple problem, i.e., classifying a class of linearly separable distributions, and show that, depen ...

2020

Optimism in the Face of Adversity: Understanding and Improving Deep Learning Through Adversarial Robustness

Pascal Frossard, Seyed Mohsen Moosavi Dezfooli, Guillermo Ortiz Jimenez, Apostolos Modas

Driven by massive amounts of data and important advances in computational resources, new deep learning systems have achieved outstanding results in a large spectrum of applications. Nevertheless, our current theoretical understanding on the mathematical fo ...

2020

Hold me tight! Influence of discriminative features on deep network boundaries

Pascal Frossard, Seyed Mohsen Moosavi Dezfooli, Guillermo Ortiz Jimenez, Apostolos Modas

Important insights towards the explainability of neural networks reside in the characteristics of their decision boundaries. In this work, we borrow tools from the field of adversarial robustness, and propose a new perspective that relates dataset features ...

2020

On The Choice of Graph Neural Network Architectures

Pascal Frossard, Guillermo Ortiz Jimenez, Clément Arthur Yvon Vignac

Seminal works on graph neural networks have primarily targeted semi-supervised node classification problems with few observed labels and high-dimensional signals. With the development of graph networks, this setup has become a de facto benchmark for a sign ...

IEEE2020

Redundant features can hurt robustness to distribution shift

Pascal Frossard, Seyed Mohsen Moosavi Dezfooli, Guillermo Ortiz Jimenez, Apostolos Modas

In this work, we borrow tools from the field of adversarial robustness, and propose a new framework that permits to relate dataset features to the distance of samples to the decision boundary. Using this framework we identify the subspace of features used ...

2020