A New PAC-Bayesian View of Domain Adaptation

Abstract : We propose a new theoretical study of domain adaptation for majority vote classifiers (from a source to a target domain). We upper bound the target risk by a trade-off between only two terms: The voters’ joint errors on the source domain, and the voters’ disagreement on the target one. Hence, this new study is simpler than other analyses that usually rely on three terms. We also derive a PAC-Bayesian generalization bound leading to a DA algorithm for linear classifiers.
Document type :
Conference papers
Complete list of metadatas

Cited literature [22 references]  Display  Hide  Download

Contributor : Emilie Morvant <>
Submitted on : Tuesday, November 3, 2015 - 5:23:46 PM
Last modification on : Thursday, July 26, 2018 - 1:10:27 AM
Long-term archiving on : Thursday, February 4, 2016 - 11:31:18 AM


Files produced by the author(s)


  • HAL Id : hal-01223164, version 1


Pascal Germain, François Laviolette, Amaury Habrard, Emilie Morvant. A New PAC-Bayesian View of Domain Adaptation. NIPS 2015 Workshop on Transfer and Multi-Task Learning: Trends and New Perspectives, Dec 2015, Montréal, Canada. ⟨hal-01223164⟩



Record views


Files downloads