Mounîm A. EL Yacoubi

CNRS, SAMOVAR,
Telecom SudParis, Paris Saclay University
FRANCE

Title:

Deep Learning: Introduction and the reasons behind its current success

Abstract:

• Introduction to Neural Nets
• Shallow Neural Networks
• Backpropagation
• Optimization of Hyper-parameters
• Types of Neural Networks
  • Multilayer Perceptrons (MLPs)
  • Convolutional Neural Nets (CNNs)
  • Autoencoders
• Introduction to Deep Learning
  • Restricted Boltzmann Machines & Deep Belief Nets (DBNs)
  • Convolutional Neural Nets (CNNs)
  • Denoising Deep Autoencoders
  • CNNs
  • Optimization of Hyper-parameters
• Example
• Conclusion

Biography:

Mounîm A. El-Yacoubi obtained a PhD in Signal Processing & Telecommunications from University of Rennes, France, in 1996. During his PhD thesis, he was with the R&D department of the Service de Recherche Technique de la Poste (SRTP) at Nantes, France, then directed by Michel Gilloux, where he developed Handwritten Address Recognition software that is still running in current French mail sorting machines, especially the first French handwritten street name recognition engine. He was a visiting scientist for 18 months at the Centre for Pattern Recognition and Machine Intelligence (CENPARMI) in Montreal, Canada, directed by Prof Ching Y. Suen, where he continued developing address recognition based on Hidden Markov Models. In 1998-2000, he become an associated professor at the Catholic University of Parana (PUC-PR) in Curitiba, Brazil, and was one of the cofounders of a research lab on document analysis in collaboration with CENPARMI and l’Ecole de Technologie Supérieure de Montréal. From 2001 to 2008, he was a Senior Scientist at Parascript, Boulder (CO, USA), a world leader company in automatic document processing, where he developed high-performance software for handwritten and machine print address, check and form recognition, still running today in Automatic reading systems in several countries across the world. Since June 2008, he is a Professor at Telecom SudParis, an engineering school, that is a member of both Institut Mines-Telecom and Paris Saclay University. His main interest lies in modeling, based on Machine Learning, human user data, especially behavioral signals like Handwriting and Gestures, with applications in e-health, human-computer interaction, and human mobility.

Mauro DALLA MURA

Grenoble INP, FRANCE
FRANCE

Title:

Feature Analysis and Unsupervised Learning

Abstract:

This course aims at giving an overview of two selected topics in Machine Learning which are feature analysis and unsupervised learning. Feature analysis comprises approaches for signal representation and discrimination based on feature space transformations. Classical methods of feature selection and extraction will be reviewed. The part related to Unsupervised learning will stem from approaches related to feature representation and review other techniques such as clustering and mixture models. Some examples of applications in remote sensing will be provided.
Outline:

• 1. Introduction (signal representation, feature space, supervised/unsupervised learning, etc)
• 2. Feature Selection (selection criteria and algorithms, measures of dissimilarity, etc)
• 3. Feature Extraction (linear and non-linear techniques, discriminant analysis, multidimensional scaling, etc)
• 4. Unsupervised learning (clustering, mixture models, hierarchical approaches, etc)

Biography:

Dr. Mauro Dalla Mura (S'08 – M'11) received the laurea (B.E.) and laurea specialistica (M.E.) degrees in Telecommunication Engineering from the University of Trento, Italy, in 2005 and 2007, respectively. He obtained in 2011 a joint Ph.D. degree in Information and Communication Technologies (Telecommunications Area) from the University of Trento, Italy and in Electrical and Computer Engineering from the University of Iceland, Iceland. In 2011 he was a Research fellow at Fondazione Bruno Kessler, Trento, Italy, conducting research on computer vision. He is currently an Assistant Professor at Grenoble Institute of Technology (Grenoble INP), France. He is conducting his research at the Grenoble Images Speech Signals and Automatics Laboratory (GIPSA-Lab). His main research activities are in the fields of remote sensing, image processing and pattern recognition. In particular, his interests include mathematical morphology, classification and multivariate data analysis. Dr. Dalla Mura was the recipient of the IEEE GRSS Second Prize in the Student Paper Competition of the 2011 IEEE IGARSS 2011 and co-recipient of the Best Paper Award of the International Journal of Image and Data Fusion for the year 2012-2013 and the Symposium Paper Award for IEEE IGARSS 2014. Dr. Dalla Mura is the President of the IEEE GRSS French Chapter since 2016 (he previously served as Secretary 2013-2016). In 2017 the IEEE GRSS French Chapter was the recipient of the IEEE GRSS Chapter Award and the “Chapter of the year 2017” from the IEEE French Section. He is on the Editorial Board of IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (J-STARS) since 2016.

Bhiksha RAJ

Carnegie Mellon University, Pittsburgh,
PA, United States

Title:

End-to-end Speech Recognition with Deep Neural Networks
Basics of Deep Learning for Speech Recognition

Abstract:

This is a three-hour tutorial, during which we will build our way up from the most basic neural network models, and work our way until a complete, simple, end-to-end automatic speech recognition system. Over the course of three hours, we will gradually build up from the simplest multi-layer perceptrons, up to convolutional neural networks, recurrent networks, and sequence-to-sequence transduction systems. We will finally converge to the three main formalisms for the compositions of complete automatic speech recognition systems using neural networks.
The topics to be covered will include:

• Basics of traditional HMM- and languge-model based speech recognition
• Basic MLP models and backpropagation for learning
• Application of MLPs to HMM-based speech recognition
• A quick intro to convolutional neural networks for feature extraction
• Recurrent networks and LSTM models
• CTC models and incorporation into language-model based recognition
• Attention models
• Attention-based end-to-end speech recognition.

It is expected that at the end of the lectures students will have a reasonably complete understanding of how state of art systems work.

Biography:

Bhiksha Raj is a professor in the School of Computer Science, at Carnegie Mellon University. Dr. Raj obtained his PhD from CMU in 2000 and was at Mistubishi Electric Research Laboratories from 2001-2008. Dr. Raj's chief research interests lie in computer audition, machine learning, deep learning, and data privacy. Dr. Raj is a fellow of the IEEE.

Ali KHENCHAF

Lab STICC UMR CNRS 6285 Laboratory,
ENSTA Bretagne, 29806, Brest,
cedex 09, FRANCE

Title:

Wave propagation, EM interaction and Remote sensing Application to the observation and extraction of parameters of the sea surface

Abstract:

In the context of the observation of the Earth's surface, remote sensing and radar imagery in the general sense make an important contribution in terms of the information collected on the areas and objects observed, and there are many different applications (territorial surveillance, cartography, oceanography, agriculture, glaciology, marine pollution, ...). This type of imaging system makes it possible, among other things, to measure the movement of the ocean surface, including the currents or wakes of ships. However, in order to contribute to the control of a situation above the surface it would be important to combine several aspects (from the sensor through the propagation medium to the processing of signals and / or images). For example, a radio link in the vicinity of the sea surface has its characteristics deeply affected by the presence of the ocean surface. Indeed, the signal from the direct path will be added a number of signals from multiple paths related to reflections from the objects and / or points of the surface. This results in interference between the direct path and the multiple paths resulting in fluctuations in the amplitude and phase of the resulting signal. These fluctuations are a function of the geometry of the link, its electromagnetic characteristics, as well as the state of the sea that depends on the weather conditions. Thus, the control of a situation above the surface is mainly through the characterization and understanding of the electromagnetic phenomena of the environment. And this is reflected directly first by the study of the propagation and interactions of electromagnetic waves with natural environments (atmosphere, rain cloud, sea, soil, forest ...) in the presence of targets or objects. Then by the control of the sensors, the understanding and realistic simulation of the monostatic or bistatic connections (of observation and / or satellites, of perception or communication) placed in this disturbed and evolutionary environment. Finally, the processing and extraction of information from a database of signals (n-D) from different sensors or after transformation constitute one of the last elements of the chain. The objective of this last element is to obtain and merge a greater amount of knowledge and information from the observed scene in order to improve the recognition and automatic identification of targets embedded in a disturbed environment. These different aspects and difficulties will be the subject of the presentation. The illustrations will be based on measurements (generated or real), the works published over several years and made in the context of different partnerships at the same time industrial, university or state.
Keywords: sensors, environment, propagation, wave interaction, radar targets, EM signature, clutter, radar imagery, detection, classification, indexing and image search, reconnaissance, remote sensing, ...

Biography:

Ali Khenchaf received the M.S. degree in statistical data processing from the University of Rennes I, Rennes, France, in 1989. In 1992, he received his Ph.D degree in Electronic Systems and Computer Network from the University of Nantes. From 1989 to 1993, he was a Researcher with the IRCCyN (UMR CNRS 6597) Laboratory, Nantes, and was an Assistant Professor from 1993 to 2001. Since September 2001, he has been with ENSTA Bretagne (Ex. ENSIETA), Brest, France, where he is currently a Professor and, from 2001 to 2011, Head of the E3I2 Laboratory (EA 3876). He joined, since January 2012, the laboratory Lab-STICC UMR CNRS 6285, where he is co-responsible of "Propagation and Interaction Multi-scales" team. The research conducted by Ali Khenchaf for over twenty five years in several laboratories are oriented towards both electromagnetic modeling and simulation, and also to the extraction and exploitation of information derived from phenomena induced by the interaction of electromagnetic waves with the environment and / or complex objects (especially sea clutter and detection problems). These activities are designed especially to integrate more "intelligence" in operational systems (airborne, satellite, drone, ...), which are dedicated to perception and observation of the natural environment. His research and teaching courses are in the fields of numerical mathematics, electromagnetic wave propagation, waves and microwave and signal processing. His research interests include radar wave scattering, microwave remote sensing, electromagnetic wave propagation, scattering in random media, monostatic and bistatic scattering of electromagnetic waves, target Radar Cross Section, Radar Imagery and target parameter estimation. He has edited or co-edited three books and author or co-authored over 300 scientific articles. He assumed responsibility of more than 40 scientific projects contracted in partnership with industry and other organizations. He also led or co-directed more than 40 students PhD thesis. Since January 2017, he has been a member of the Editorial Board of the European Journal of Remote Sensing. And professor Khenchaf is currently guest editor of the journal "Remote sensing". In addition, Ali Khenchaf is expert with several agencies and organizations in France and abroad.

Mehrez ZRIBI

CESBIO, FRANCE
FRANCE

Title:

Abstract:

Biography:

M. Zribi is a Research Director with Centre National de Recherche Scientifique (CNRS). He received the B.E. degree in signal processing from the Ecole Nationale Supérieure d’Ingénieurs en Constructions Aéronautiques, Toulouse, France, and the Ph.D. degree from the Université Paul Sabatier, Toulouse. In 1995, he joined the Centre d’Etude des Environnements Terrestre et Planétaires Laboratory/Institut Pierre Simon Laplace, Vélizy, France. In 2001, he joined CNRS organism. Since October 2008, he has been with the Centre d’Etudes Spatiales de la Biosphère (CESBIO), Toulouse. He is responsible of the team of observation systems in CESBIO.
His research interests include microwave remote sensing applied to hydrology, microwave modelling for land surface parameters estimations and finally airborne microwave instrumentation. He has acted as project manager of many projects supported by the French Space Agency (CNES), French National Program of Remote Sensing, the European Space Agency (ESA), French National Research Agency (ANR)…. He has published more than 100 articles in refereed journals and more than 200 conference proceedings. He has published twenty books about remote sensing theory and applications. His H index=30 (WOS). He is member of comities for airborne measurements (French CSTA, European EUFAR). He is in editor boards of three journals: Journal of Geophysical Instrumentation, Methods and data systems (EGU), Sensors/MDPI and Nature/Scientific reports. He is senior IEEE member.

J. Philippe GASTELLU-ETCHERGORRY

CESBIO, FRANCE
FRANCE

Title:

Abstract:

Biography:

Dijana Petrovska Delacrétaz

Mines Télécom -Télécom SudParis, FRANCE
FRANCE

Title:

Deep Neural Nets and Face Recognition

Abstract:

Biography:

Riadh Ksontini

SUP’COM, Tunis
TUNISIA

Title:

ILinear Models and Deep Learning for Data Visualization

Abstract:

In Direct Volume Rendering (DVR), the Transfer Function (TF) to map voxel values to color and opacity values is difficult to obtain. Existing TF design tools are complex and non-intuitive for the end user, who is more likely to be a medical professional than an expert in image processing or data modeling. We propose a novel image-centric volume visualization method where the user directly works on the volume data to simply select the parts he/she would like to visualize. The user’s work is further simplified by presenting only the most informative volume slices for selection. Based on the selected parts, all the voxels are classified using our Sparse Nonparametric Support Vector Machine (SN-SVM) classifier and deepl learning methods, which combine both local and near-global distributional information of the training data to obtain accurate results. The voxel classes are then mapped to color and opacity values using the concept of harmonic colors, which provides easily distinguishable and aesthetically pleasing results. The need for visualization emerged from the necessity of informative exploration of data in several domains, such as, medical imaging, seismology, computer network analysis, traffic analysis ans organizational management. For example, in medical visualization, Magnetic Resonance Imaging (MRI) or Computed Tomography (CT) scan data shows the visual representation of the patient’s inner organs with an overlay of necessary information (e.g.intensity of the voxels).

Biography:

Heni BOUHAMED

FSEGS, Sfax
TUNISIA

Title:

Deep neural network et applications (python avec Keras, Tensorflow etH2O)

Plan:

• - Exemple introductif avec la base de données benchmark IRIS
• - Classification d’images avec le « Local Binary Pattern »
  → Application sur une base de données pour la reconnaissance de Piéton/voiture/moto.
• - Classification en biologie/génétique
  → Application sur une base de données pour la classification de gènes selon leurs implications aux génodermatoses.

Bibliographie :

• Sujit Pal, Antonio Gulli, Deep Learning withKeras, PacktPublishing Ltd, 26 avr. 2017 - 318 pages
• Darren Cook, Practical Machine Learning with H2O, Published by O’Reilly Media, Inc., 1005 Gravenstein Highway North, Sebastopol, CA 95472. Edition 2016.
• Jason Brownlee, Deep Learning With Python, Copyright 2016 Jason Brownlee. Edition: v1.7
• F.Dornaika, A.Bosaghzadeh, H.Salmane, Y.Ruichek, A graph construction methodusing LBP self-representativeness for outdoorobjectcategorization, Engineering ApplicationsofArtificial Intelligence36(2014)294–302.
• F. Dornaika, A. Bosaghzadeh, H. Salmane, Y. Ruichek, Graph-based semi-supervisedlearningwith Local Binary Patterns for holisticobjectcategorization, Expert Systemswith Applications 41 (2014) 7744–7753
• Mariem kchaou, Lilia Romdhane, Heni Bouhamed. « ReducedalgorithmiccomplexitywhenlearningBayesian classifier structure using score-basedalgorithms », The 5th International Conference on Control Engineering &Information Technology (CEIT-2017)

Biography: