Generating Relevant Counter-Examples from a Positive Unlabeled Dataset for Image Classification

Florent Chiaroni (1, 2) , Ghazaleh Khodabandelou (3) , Mohamed-Cherif Rahal (2) , Nicolas Hueber (4) , Frédéric Dufaux (1)

1 L2S - Laboratoire des signaux et systèmes
2 Equipe Perception - Département de conduite déléguée VEH08 [Versailles]
3 Equipe Intelligence Artificielle - Département des nouvelles utilisations MOB05 [Versailles]
4 ISL - Institut franco-allemand de recherches de Saint-Louis

With surge of available but unlabeled data, Positive Unlabeled (PU) learning is becoming a thriving challenge. This work deals with this demanding task for which recent GAN-based PU approaches have demonstrated promising results. Generative adversarial Networks (GANs) are not hampered by deterministic bias or need for specific dimensionality. However, existing GAN-based PU approaches also present some drawbacks such as sensitive dependence to prior knowledge, a cumbersome architecture or first-stage overfitting. To settle these issues, we propose to incorporate a biased PU risk within the standard GAN discriminator loss function. In this manner, the discriminator is constrained to request the generator to converge towards the unlabeled samples distribution while diverging from the positive samples distribution. This enables the proposed model, referred to as D-GAN, to exclusively learn the counter-examples distribution without prior knowledge. Experiments demonstrate that our approach outperforms state-of-the-art PU methods without prior by overcoming their issues.

Mots clés: generative models; semi-supervised learning; Generative Adversarial Networks (GANs); partially supervised learning; Image classification; deep learning

Domaines: Machine Learning [stat.ML]; Apprentissage [cs.LG]; Réseau de neurones [cs.NE]; Intelligence artificielle [cs.AI]; Vision par ordinateur et reconnaissance de formes [cs.CV]; Traitement du signal et de l'image [eess.SP]; Traitement des images [eess.IV]; Informatique

Florent Chiaroni, Ghazaleh Khodabandelou, Mohamed-Cherif Rahal, Nicolas Hueber, Frédéric Dufaux. Generating Relevant Counter-Examples from a Positive Unlabeled Dataset for Image Classification. Pattern Recognition, 2020, 107, pp.107527. ⟨10.1016/j.patcog.2020.107527⟩. ⟨hal-02302682v1⟩