Sonia Tabti

Sonia Tabti
Fieldbox

PhD
Define and supervise research activities applied to various industries. From time series to image modeling.

About

13
Publications
1,278
Reads
How we measure 'reads'
A 'read' is counted each time someone views a publication summary (such as the title, abstract, and list of authors), clicks on a figure, or views or downloads the full-text. Learn more
172
Citations
Additional affiliations
February 2019 - February 2019
Fieldbox.ai
Position
  • Analyst
Description
  • I am working on different R&D topics for industrial applications and mostly on time series forecasting, classification and signal processing.
January 2017 - July 2018
Université de Caen Normandie
Position
  • PostDoc Position
Description
  • I developed variants of TV defined on graphs in order to perform image and 3D point clouds filtering and segmentation.
September 2016 - December 2016
University of Naples Federico II
Position
  • PostDoc Position
Education
September 2011 - September 2012
Ecole normale supérieure de Cachan
Field of study
  • Mathematics vision and statistical learning

Publications

Publications (13)
Conference Paper
Full-text available
Cet article explore l'utilisation d'AutoEncodeurs Variationnels (VAE) parcimonieux dans le cadre de l'analyse des perturbations affectant la distribution de données industrielles, aussi appelées shifts. À cette fin, plusieurs modèles sont comparés, en particulier, nous introduisons le LassoVAE, un VAE avec dé-codeur parcimonieux dont la procédure d...
Conference Paper
Full-text available
How relevant are interpretability methods designed for deep learning models in the context of visual defect detection? Beyond their actual output, to what extent can these methods be used in a production environment? We study and evaluate interpretability methods for convolutional neural networks (CNN) and vision transformers (ViT) on image classif...
Conference Paper
Full-text available
In this article, a computationally efficient manifold learning algorithm combining a variational auto-encoder and a nearest neighbor graph is proposed. In fact, using a variational autoencoder to compute an approximation of the underlying data distribution allows our method to tackle some shortcomings of neighbor graph construction methods, namely...
Article
Full-text available
In this paper, we propose a sparsity-based despeckling ap- proach. The first main contribution of this work is the elabo- ration of a sparse-coding algorithm adapted to the statistics of SAR images. In fact, most sparse-coding algorithms for SAR data apply a logarithmic transform to data, so as to convert the noise from multiplicative to additive....
Article
Speckle reduction is a longstanding topic in synthetic aperture radar (SAR) imaging. Since most current and planned SAR imaging satellites operate in polarimetric, interferometric or tomographic modes, SAR images are multi-channel and speckle reduction techniques must jointly process all channels to recover polarimetric and interferometric informat...
Thesis
RESUME : Le traitement des images satéllitaire est un enjeu important dans notre société où l’on a besoin de surveiller la planète pour évaluer la déforestation ou encore réaliser le bilan d’une catastrophe naturelle. Les images radarsont utiles pour cela car elles sont obtenues par un système actif et donc indépendamment de l’éclairement solaire e...
Conference Paper
Full-text available
Adding invariance properties to a dictionary-based model is a convenient way to reach a high representation capacity while maintaining a compact structure. Compact dictionaries of patches are desirable because they ease semantic interpretation of their elements (atoms) and offer robust decompositions even under strong speckle fluctuations. This pap...
Conference Paper
Full-text available
Patches have proven to be very effective features to model natural images and to design image restoration methods. Given the huge diversity of patches found in images, modeling the distribution of patches is a difficult task. Rather than attempting to accurately model all patches of the image, we advocate that it is sufficient that all pixels of th...

Network

Cited By