Multi-class to Binary reduction of Large-scale classification Problems

Bikash Joshi (1) , Massih-Reza Amini (1) , Ioannis Partalas (2) , Liva Ralaivola (3) , Nicolas Usunier (4) , Gaussier Eric (1)

1 AMA - Analyse de données, Modélisation et Apprentissage automatique [Grenoble]
2 VISEO
3 QARMA - éQuipe AppRentissage et MultimediA [Marseille]
4 FAIR - Facebook AI Research [Paris]

Large-scale multi-class classification problems have gained increased popularity in recent time mainly because of the overwhelming growth of textual and visual data in the web. However, this is a challenging task for many reasons. The main challenges in Large-scale classification problems are: scalability, complexity of model and class imbalance problem. In this work, we present an algorithm for binary reduction of multi-class classification problems, which aims at addressing the above-mentioned challenges.

Mots clés: Large scale multiclass classification; reduction

Domaines: Apprentissage [cs.LG]

Bikash Joshi, Massih-Reza Amini, Ioannis Partalas, Liva Ralaivola, Nicolas Usunier, et al.. Multi-class to Binary reduction of Large-scale classification Problems. International Workshop on Big Multi-Target Prediction ECML/PKDD 2015, Sep 2015, Poto, Portugal. ⟨hal-01294379⟩