Wasserstein Barycenter Model Ensembling
By Pierre L. Dognin, Igor Melnyk, Youssef Mroueh, Jerret Ross, C. D. Santos, Tom Sercu. 2019
27 citations recorded by the source corpus.
In this paper we propose to perform model ensembling in a multiclass or a multilabel learning setting using Wasserstein (W.) barycenters. Optimal transport metrics, such as the Wasserstein distance, allow incorporating semantic side information such as word embeddings. Using W. barycenters to find the consensus between models allows us to balance confidence and semantics in finding the agreement between the models. We show applications of Wasserstein ensembling in attribute-based classification, multilabel learning and image captioning generation. These results show that the W. ensembling is a viable alternative to the basic geometric or arithmetic mean ensembling.
Subfields: Artificial Intelligence, Control and Systems Engineering
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Carlo Ciliberto
- Massimiliano Pontil
- Saverio Salzo
- Joaquín Fontbona
- Felipe A. Tobar
- Gonzalo Rios
- Giulia Luise
- Julio D. Backhoff Veraguas
In Control and Systems Engineering
- Massimiliano Pontil
- Carlo Ciliberto
- Ding-Xuan Zhou
- Lorenzo Rosasco
- Andrea Caponnetto
- Steve Smale
- Ernesto De Vito
Citations in the corpus
- Bayesian Learning with Wasserstein Barycenters (2018, methodology): “The reader is referred to [36, 14, 29, 16] for alternative approaches to computing Wasserstein barycenters.”
- Sinkhorn Barycenters with Free Support via Frank-Wolfe Algorithm (2019, methodology): “In particular, Wasserstein barycenters have been successfully used in settings such as texture mixing [40], Bayesian inference [49], imaging [26], or model ensemble [18].”