Robust deep learning: A case study

Victor Estrade 1, 2, 3 Cécile Germain 1, 2, 3 Isabelle Guyon 1, 2, 3, 4 David Rousseau 5
1 TAU - TAckling the Underspecified
LRI - Laboratoire de Recherche en Informatique, Inria Saclay - Ile de France, CNRS - Centre National de la Recherche Scientifique : UMR8623, UP11 - Université Paris-Sud - Paris 11
Abstract : We report on an experiment on robust classification. The literature proposes adversarial and generative learning, as well as feature construction with auto-encoders. In both cases, the context is domain-knowledge-free performance. As a consequence, the robustness quality relies on the representativity of the training dataset wrt the possible perturbations. When domain-specific a priori knowledge is available, as in our case, a specific flavor of DNN called Tangent Propagation is an effective and less data-intensive alternative.
Document type :
Documents associated with scientific events
Complete list of metadatas

Cited literature [7 references]  Display  Hide  Download
Contributor : Cecile Germain <>
Submitted on : Sunday, December 17, 2017 - 6:17:10 PM
Last modification on : Wednesday, January 23, 2019 - 2:39:30 PM


Files produced by the author(s)


  • HAL Id : hal-01665938, version 1


Victor Estrade, Cécile Germain, Isabelle Guyon, David Rousseau. Robust deep learning: A case study. JDSE 2017 - 2nd Junior Conference on Data Science and Engineering, Sep 2017, Orsay, France. pp.1-5, 2017. ⟨hal-01665938⟩



Record views


Files downloads