February, the 14th, 2014 – Kick Off
This meeting took place in the Hubert Curien Lab in Saint-Etienne.
Presents |
|
|
|
Schedule |
09h40 – 10h00 : Arrival, Introduction |
10h00 – 10h50 : Taygun Kekec Spatio-Temporal Scene Understanding using Deep Learning |
10h50 – 11h40 : Stefan Duffner Face Analysis in Images and Videos |
11h40 – 12h30 : Remi Emonet Temporal Probabilistic Models |
14h00 – 14h50 : Christine Solnon Overview of the results of the SATTIC ANR project |
15h10 – 16h30 : Elisa Fromont (Reminder of the tasks for the SoLSTiCe and discuss about internship, thesis, logistics, demonstrations, website, etc.) Enter password |
June, 19th & 20th, 2014 – Semi Annual Meeting
This meeting took place in Goutlelas’ castle.
More information about this place here : http://www.chateaudegoutelas.fr/.
Presents |
|
|
|
Schedule |
Thursday, June 19th, 2014 |
09h40 – 10h00 : Arrival, Introduction |
10h30 – 11h30 : Guillaume Damiand Remove noise in video with 3D topological Enter password |
11h30 – 12h30 : Samba Ndojh Ndiaye Maximum common subgraphs Enter password |
14h00 – 14h45 : Christian Wolf Structuration of the deep learning |
14h45 – 16h15 : Elisa, Rémi & Christian’s presentation about Taygun’s work (CAP2014+ BMVC2014). Presentation of one or two state-of-the-art articles on this subject (s.t. Scene parsing by integrating function, geometry and appearance model, CVPR 2013, Indor semantic segmentation using depth information, ICLR 2013…) in order to discuss the PhD topic of Damien Fourure Enter password |
16h30 – 17h15 : Discussion about the third thesis. Presentation of the ICCV2013 article (from Marc, Amaury, Basura) and Rahaf’s works. |
17h30 – 19h00 : Walk through the volcanic trail through the Castle. |
Friday, June 20th, 2014 |
09h00 – 10h15 : Rémi Emonet Tutorial about the Fisher Vectors and presentation of Valentina’s work (Master 1). Présentation : http://twitwi.github.io/Presentation-2014-solstice-fisher/ |
10h30 – 11h30 : Romain Deville (Master 2) Fouille de Grilles pour l’analyse d’Images Enter password |
11h30 – 12h30 : Amaury Habard, Papers CAP2014 + ECCV2014 (with Marc, Damien and Michaël) and Michaël’s thesis subject. |
14h00 – 15h00 : Baptiste Jeudy presentation about mining of the greates tiles in streams, application to video (ECML/PKDD 2014). Enter password |
15h00 – 16h00 : Miscellaneous talk on project. |
March, 6th, 2015 – Semi Annual Meeting
This meeting took place in the Hubert Curien Lab in Saint-Etienne.
Presents |
|
|
|
Schedule |
09h30 – 10h00 : Arrival, Introduction |
10h00 – 11h00 : Stefan Duffner Exploiting contextual motion cues for visual object tracking |
11h00 – 12h00 : Damien Fourrure Deep Color Constancy |
12h00 – 12h45 : Alain Tremeau. |
13h30 – 14h30h : Romain Deville Classification with Bag of Grids, progression and last results. |
14h30 – 15h00 : Christine Solnon |
15h00 – 15h30 : Divers talk on project |
November, 26th, 2015 – Saint&Lyon Deep Learning Workshop
This meeting took place in Lyon, in the campus « La Doua ».
You can find the program other information on the workshop website : http://perso.univ-st-etienne.fr/fod07375/Workshop/ .
Presents |
|
|
|
Schedule |
09h30 – 10h00 : Arrival, Welcoming |
10h00-10h45 : Damien Fourure
What is a Neural Network ? A quick introduction.Transferability.
|
10h45-11h00 : Coffee breakSponsored by Solstice ANR Project |
11h00-11h45 : Bastien Moysset
Object Detection with Recurrent Neural Networks.
|
11h45-13h30 : Lunch |
13h30-14h15 : Christian Wold
DeepLeanrning vs Random Forests
|
14h15-15h00 : Mohamed Elawady
|
15h00-15h30 : Coffee breakSponsored by Solstice ANR Project |
15h30-16h15 : Natalia Neverova
|
16h15-17h00 : Rémi Emonet
|
17h00-17h30 : Debriefing |
January, 21st and 22nd, 2016 – Semi Annual Meeting
This meeting took place in Chalmazel le Bourg.
Presents |
|
|
|
Schedule |
Thusrday, January 21st, 2016 |
10h00 – 10h30 : Arrival, Welcoming |
10h30-11h30 : Marc Sebban Parsimony in Convex Optimization for Supervised Machine Learning |
12h15-14h00 : Lunch Break at Ferme Auberge des Granges |
14h00-15h00 : Ducottet Christophe String representations and distances in deep convolutional neural networks for image classification |
15h00-16h00 : Muselet Damien Learning to rank based on subsequences, ICCV 2015 |
16h00-16h30 Break |
16h30-17h30 : Wolf Christian Segmentation with topological constraint by shortest path search in a graph |
17h30-18h30 : Fourrure Damien Semantic Scene Parsing Using Inconsistent Labellings |
Thusrday, January 21st, 2016 |
8h45-9h45 : Salotti Julien Spatio-temporal analysis of traffic Data for smart mobility |
9h45-10h45 : Minot Maël Maximum common subgraph |
10h45-11h15 : Break |
11h15-12h15 : Deville Romain Bag of grid for image classification |
12h15-16h15 : Lunch Break at Ferme Auberge des granges |
16h15-17h15 : Solnon Christine Machine learning for graph matching |
16h15-17h15 : Rémi Emonet Tutorial on gaussian process. |
January, 4th, 2017 – Saint&Lyon Deep Learning Workshop
This meeting took place in Saint-Etienne, in the campus « Manufacture ».
You can find the program other information on the workshop website : http://perso.univ-st-etienne.fr/fod07375/Saint&Lyon/ .
The morning and afternoon coffee breaks were sponsored by the Solstice ANR project.
PresentsThe workshop hosted 40 participants from Lyon and Saint-Etienne. The list of the participants can be found here. |
Schedule |
9h15-12h15 : Damien Fourure
Tutorial on deep learning using TORCH 7
|
13h30-14h15 : Bastien Moysset
Deep learning for detection and localization of objects in imagesThe current trend in object detection and localization is to learn predictions with high capacity deep neural networks. In this talk, the main state of the art techniques will be presented (RCNN, Fast RCNN, Faster RCNN, MultiBox, YOLO et SSD) along with our technique designed for the detection of many objects from fewer examples. We will emphasize on the similarities and differences between these techniques and we will try to list their pros and cons. |
14h15-15h00 : Julien Tissier
Natural Language Processing using Neural NetworksNatural Language Processing (NLP) is a subfield of artificial intelligence and computational linguistics. The main objective is to teach the machine the human language. To learn the language and solve some NLP tasks, machines need a representation for words. My seminar will mainly focus on how to learn a good word representation that carry both semantic and syntactic information by using neural networks. I will then present the method I developed during my internship to improve the semantic similarity between words, by using online lexical dictionaries. |
15h15-16h00 : Christian Wolf
Graphical Models and Deep NetworksWe first present a very brief introduction into graphical models and their principal inference algorithms. We then present the differences with deep networks and give concrete examples for each family (deformable parts models, kinematic trees, attention mechanisms etc.). We will finish with some parallels to human psychology and human thinking. |
16h15-17h00 : Christophe Ducottet
String representations and distances in deep convolutional neural networks for image classificationRecent advances in image classification mostly rely on the use of powerful local features combined with an adapted image representation. Although Convolutional Neural Network (CNN) features learned from ImageNet were shown to be generic and very efficient, they still lack of flexibility to take into account variations in the spatial layout of visual elements. In this paper, we investigate the use of structural representations on top of pretrained CNN features to improve image classification. Images are represented as strings of CNN features. Similarities between such representations are computed using two new edit distance variants adapted to the image classification domain. Our algorithms have been implemented and tested on several challenging datasets, 15Scenes, Caltech101, Pascal VOC 2007 and MIT indoor. The results show that our idea of using structural string representations and distances clearly improves the classification performance over standard approaches based on CNN and SVM with linear kernel, as well as other recognized methods of the literature. |
17h00-17h45 : Remi Emonet
Likelihood-based and Likelihood free Unsupervised Learning |
June, 27 2017 – Semi Annual Meeting
This meeting took place in Côte Claire domains at Saint-Sorlin-de-Morestel
Presents |
|
|
|
Monday June 26th, 2017 |
Schedule |
10h00 – 10h30 : Arrival, Introduction and Bilan |
10h30 – 10h50 : Coffee break |
10h50 – 11h30 : Julien Salotti : Comment l’information causale améliore-t-elle la prévision à court-terme du trafic urbain ? Une approche portfolio. Enter password |
11h30 – 12h10 : Romain Deville : Mining patterns in spatio-temporal grids |
14h00 – 14h40 : Maël Minot : Combining constraint programming and linear programming to solve the sum colouring problem. Enter password |
14h40 – 15h20 : Kevin Bascol : Deep Learning for anomaly detection in Chair-Lifts videos. Enter password |
15h20 – 15h40 : Coffee break |
15h40 – 16h20 : Rémi Émonet : Landmark-based Linear Local Support Vector Machines. Enter password |
16h20 – 17h00 : Christophe Ducottet : Exploring Global Reflection Symmetry in Photographs & Visual Arts Enter password |
Tuesday June 27th, 2017 |
9h00 – 9h40 : Christian Wolf : What we are up to : Current work on human action recognition using recurrent network. |
9h40 – 10h30 : Damien Fourrure : GridNet une architecture spécialisée pour la segmentation sémantique Enter password |
10h30 – 10h50 : Coffee break |
10h50 – 11h30 : Marc Sebban : Learning Maximum Excluding Ellipsoids In Unbalanced Scenarios With Theoretical Guarantees Enter password Enter password |
11h30 – 12h10 : Stefan Duffner : Apprentissage de similarités avec réseaux de neurones Enter password |
14h00 – 14h40 : Damien Muselet : 3D color charts for computer vision. Enter password |
14h40 – 15h00 : Conclusion & Future Work |