Putting theory to work: from learning bounds to meta-learning algorithms

Quentin Bouniot (1, 2) , Ievgen Redko (2) , Romaric Audigier (1) , Angelique Loesch (3) , Amaury Habrard (2)

1 LIST (CEA) - Laboratoire d'Intégration des Systèmes et des Technologies
2 LHC - Laboratoire Hubert Curien
3 LVIC - Laboratoire Vision et Ingénierie des Contenus

In this paper, we review the recent advances in meta-learning theory and show how they can be used in practice both to better understand the behavior of popular meta-learning algorithms and to improve their generalization capacity. This latter is achieved by integrating the theoretical assumptions ensuring efficient meta-learning in the form of regularization terms into several popular meta-learning algorithms for which we provide a large study of their behavior on classic few-shot classification benchmarks. To the best of our knowledge, this is the first contribution that puts the most recent learning bounds of meta-learning theory into practice for the popular task of few-shot classification.

Domaines: Intelligence artificielle [cs.AI]; Vision par ordinateur et reconnaissance de formes [cs.CV]; Apprentissage [cs.LG]; Traitement des images [eess.IV]

Quentin Bouniot, Ievgen Redko, Romaric Audigier, Angelique Loesch, Amaury Habrard. Putting theory to work: from learning bounds to meta-learning algorithms. Workshop Meta-Learn@NeurIPS 2020, Dec 2020, Vancouver (Virtual conference), Canada. ⟨cea-03251827⟩