Skip to Main content Skip to Navigation
Conference papers

Performances optimales pour les données multimodales partiellement couplées

Abstract : Two models are called " coupled " when a non empty set of the underlying parameters of interest are related through a differentiable implicit function. In this context, several estimation strategies can be derived: either a joint estimation where the parameters of interest are retrieved by merging all datasets, or an individual estimation where the parameters of interest are respectively estimated from each dataset. In this paper, we show by analyzing lower bounds that the optimal process of joint estimation is always more accurate than the individual estimation process in the sense of mean square error for a general class of dataset distributions. This property is still true for the parameters that are not directly linked. Finally, we illustrate our results with the fusion of multiple tensor data.
Complete list of metadatas

Cited literature [19 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-01567124
Contributor : Pierre Comon <>
Submitted on : Friday, July 21, 2017 - 5:33:36 PM
Last modification on : Tuesday, May 26, 2020 - 6:50:29 PM

File

ren52.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01567124, version 1

Citation

Chengfang Ren, Rodrigo Cabral Farias, Pierre-Olivier Amblard, Pierre Comon. Performances optimales pour les données multimodales partiellement couplées. XXVIème colloque GRETSI (GRETSI 2017), Sep 2017, Juan-Les-Pins, France. ⟨hal-01567124⟩

Share

Metrics

Record views

446

Files downloads

95