Robust Domain Adaptation: Representations, Weights and Inductive Bias - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2020

Robust Domain Adaptation: Representations, Weights and Inductive Bias

Philippe Very
  • Fonction : Auteur
  • PersonId : 1085227
Clément Chastagnol
  • Fonction : Auteur
  • PersonId : 1085228

Résumé

Unsupervised Domain Adaptation (UDA) has attracted a lot of attention the past ten years. The emergence of Domain Invariant Representations (IR) has improved drastically the transferability of representations from a labelled source domain to a new and unlabelled target domain. However, a potential pitfall of this approach, e.g. the presence of label shift, has been brought to light. Some works address this issue with a relaxed version of domain invariance obtained by weighting samples, a strategy often referred to as Importance Sampling. From our point of view, the theoretical aspects of how Importance Sampling and Invariant Representations interact in UDA have not been studied in depth. In the present work, we present a bound of the target risk which incorporates both weights and invariant representations. Our theoretical analysis highlights the role of inductive bias in aligning distributions across domains. We illustrate it on standard benchmarks by proposing a new learning procedure for UDA. We observed empirically that weak inductive bias makes adaptation more robust. The elaboration of stronger inductive bias is a promising direction for new UDA algorithms.
Fichier principal
Vignette du fichier
Robust-Domain-adaptation-representations-weights-inductive-bias.pdf (673.41 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03050623 , version 2 (10-12-2020)
hal-03050623 , version 1 (06-12-2023)

Identifiants

Citer

Victor Bouvier, Philippe Very, Clément Chastagnol, Myriam Tami, Céline Hudelot. Robust Domain Adaptation: Representations, Weights and Inductive Bias. ECML 2020, Sep 2020, Gand ( virtuel ), Belgium. ⟨10.1007/978-3-030-67658-2_21⟩. ⟨hal-03050623v2⟩
85 Consultations
100 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More