Multi-Domain Adversarial Learning

Alice Schoenauer Sebag 1 Louise Heinrich 1 Marc Schoenauer 2 Michèle Sebag 3, 2 Lani Wu 1 Steven Altschuler 1
2 TAU - TAckling the Underspecified
LRI - Laboratoire de Recherche en Informatique, UP11 - Université Paris-Sud - Paris 11, Inria Saclay - Ile de France, CNRS - Centre National de la Recherche Scientifique : UMR8623
Abstract : Multi-domain learning (MDL) aims at obtaining a model with minimal average risk across multiple domains. Our empirical motivation is automated microscopy data, where cultured cells are imaged after being exposed to known and unknown chemical perturbations, and each dataset displays significant experimental bias. This paper presents a multi-domain adversarial learning approach, MULANN, to leverage multiple datasets with overlapping but distinct class sets, in a semi-supervised setting. Our contributions include: i) a bound on the average-and worst-domain risk in MDL, obtained using the H-divergence; ii) a new loss to accommodate semi-supervised multi-domain learning and domain adaptation; iii) the experimental validation of the approach, improving on the state-of-the-art on two standard image benchmarks, and a novel bioimage dataset, CELL.
Document type :
Conference papers
Complete list of metadatas

Cited literature [20 references]  Display  Hide  Download

https://hal.inria.fr/hal-01968180
Contributor : Marc Schoenauer <>
Submitted on : Tuesday, January 8, 2019 - 12:24:07 AM
Last modification on : Friday, May 10, 2019 - 1:08:06 PM
Long-term archiving on : Tuesday, April 9, 2019 - 1:09:16 PM

Identifiers

  • HAL Id : hal-01968180, version 1

Citation

Alice Schoenauer Sebag, Louise Heinrich, Marc Schoenauer, Michèle Sebag, Lani Wu, et al.. Multi-Domain Adversarial Learning. ICLR 2019 - Seventh annual International Conference on Learning Representations, Tara Sainath, May 2019, New Orleans, United States. ⟨hal-01968180⟩

Share

Metrics

Record views

345

Files downloads

605