Éditeur(s) : HAL CCSD Institute of Electrical and Electronics Engineers
Résumé : International audience
This paper presents our response to the first interna- tional challenge on Facial Emotion Recognition and Analysis. We propose to combine different types of features to automatically detect Action Units in facial images. We use one multi-kernel SVM for each Action Unit we want to detect. The first kernel matrix is computed using Local Gabor Binary Pattern histograms and a histogram intersection kernel. The second kernel matrix is computed from AAM coefficients and an RBF kernel. During the training step, we combine these two types of features using the recently proposed SimpleMKL algorithm. SVM outputs are then averaged to exploit temporal information in the sequence. To eval- uate our system, we perform deep experimentations on several key issues: influence of features and kernel function in histogram- based SVM approaches, influence of spatially-independent in- formation versus geometric local appearance information and benefits of combining both, sensitivity to training data and interest of temporal context adaptation. We also compare our results to those of the other participants and try to explain why our method had the best performance during the FERA challenge.
ISSN: 1083-4419

hal-00731864
http://hal.upmc.fr/hal-00731864
http://hal.upmc.fr/hal-00731864/document
http://hal.upmc.fr/hal-00731864/file/senechal12tsmc_preprint.pdf
DOI : 10.1109/TSMCB.2012.2193567

Éditeur(s) : HAL CCSD
Résumé : International audience
We instrumented a realistic car simulator to extract low level data related to the driver's use of the vehicle controls. After proceeding these data, we generated features that were fed to a Multi-Layer Perceptron (MLP) and Support Vector Machines (SVM). Our goal was determine if the driver's Blood Alcohol Content (BAC) was over $0.4g.l^{-1}$ or not, and even estimate the BAC value. Our device process the vehicle's controls data and then outputs the user BAC. We discuss the results of the prototype using the MLP and SVM algorithms in both single-user and multi-user context for detection of drunk drivers and estimation of the BAC value. The prototype performed better with single user base than with multi-user, and provided comparable results with MLP and SVM.
ESANN 2013 proccedings,European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning
Bruges, Belgium

hal-00850138
https://hal.archives-ouvertes.fr/hal-00850138

Éditeur(s) : HAL CCSD
Résumé : International audience
Extract the features for a multi-touch gesture is difficult due to the complex temporal and motion relations between multiple trajectories. In this paper we present a new generic graph model to quantify the shape, temporal and motion information from multi-touch gesture. To make a comparison between graph, we also propose a specific graph matching method based on graph edit distance. Results prove that our graph model can be fruitfully used for multi-touch gesture pattern recognition purpose with the classifier of graph embedding and SVM.
17th Biennial Conference of the International Graphonomics Society
Pointe-à-Pitre, Guadeloupe

hal-01165768
https://hal.univ-antilles.fr/hal-01165768
https://hal.univ-antilles.fr/hal-01165768/document
https://hal.univ-antilles.fr/hal-01165768/file/IGS_2015_submission_6.pdf

Éditeur(s) : HAL CCSD Wiley
Résumé : International audience
A relevant and accurate description of three-dimensional (3D) protein structures can be achieved by characterizing recurrent local structures. In a previous study, we developed a library of 120 3D structural prototypes encompassing all known 11-residues long local protein structures and ensuring a good quality of structural approximation. A local structure prediction method was also proposed. Here, overlapping properties of local protein structures in global ones are taken into account to characterize frequent local networks. At the same time, we propose a new long local structure prediction strategy which involves the use of evolutionary information coupled with Support Vector Machines (SVMs). Our prediction is evaluated by a stringent geometrical assessment. Every local structure prediction with a Calpha RMSD less than 2.5 A from the true local structure is considered as correct. A global prediction rate of 63.1% is then reached, corresponding to an improvement of 7.7 points compared with the previous strategy. In the same way, the prediction of 88.33% of the 120 structural classes is improved with 8.65% mean gain. 85.33% of proteins have better prediction results with a 9.43% average gain. An analysis of prediction rate per local network also supports the global improvement and gives insights into the potential of our method for predicting super local structures. Moreover, a confidence index for the direct estimation of prediction quality is proposed. Finally, our method is proved to be very competitive with cutting-edge strategies encompassing three categories of local structure predictions. Proteins 2009. (c) 2009 Wiley-Liss, Inc.
ISSN: 0887-3585

inserm-00348740
http://www.hal.inserm.fr/inserm-00348740
http://www.hal.inserm.fr/inserm-00348740/document
http://www.hal.inserm.fr/inserm-00348740/file/Bornot_Proteins_2009_preprint_version.pdf
http://www.hal.inserm.fr/inserm-00348740/file/inserm-00348740_edited.pdf