Professor, Polytech Annecy-Chambéry.
Contact.
Email : alexandre.benoit –@– univ-smb.fr
Phone : +33(0) 4 50 09 65 22 – télécopie : +33(0) 4 50 09 65 22
Office : A223
Postal address : LISTIC – Polytech Annecy Chambéry, BP 80439, 74944 Annecy le Vieux Cedex, France
Position / responsabilities.
- Permanent LISTIC Lab. member, professor position
- Main teaching activities at Polytech Annecy Chambéry, Université Savoie Mont Blanc.
Research interests.
Member of the “Apprentissage, Fusion et Télédétection (AFuTé) thematic” with strong connexions with ReGaRD thematics on distributed systems, HPC, data and ethics,
- Image, video and more generally data understanding : from data preprocessing, features extraction to concept recognition (objects, action, etc.) to classical 1D sensors.
- Heterogeneous Data Fusion : associating multimodal inputs, images, contextual indicators, and others to make the decision process more reliable.
- Sensor time series in general, from image sequences (multimedia, satellite remote sensing, astrophysics imaging) to classical sensors analysis.
- Current methodological approaches rely on deep learning methods with some specific directions:
- Frugal models, low computational cost by design: light architectures efficient in terms of training efficacy as well as reduced inference time.
- Explainable models, with two approaches:
- physical aware models design : by introducing physical knowledge in the model structure or by introducing optimization criteria related to the physical phenomenon to be considered.
- developing XAI visualization techniques to help explain and understand model behaviors and predictions.
- Physically informed models, ongoing activity
- Federated learning, distributed and privacy preserving machine learning, ongoing activity.
Keywords
Video clustering Classification Retina modelDeep Learning Semi-supervised learning 3D convolutional neural network V1 cortex model Semantic Segmentation Real-time retina Human vision Remote sensingSpectral Clustering 2D metrics 3DOpenInterface Deep learning Attention level, Pairwise constraints Graph Cut
Publications
An overview of my publications is provided here : https://cv.archives-ouvertes.fr/alexandre-benoit and here also (subject to bugs) : registered in HAL
Research Projects
- A brief overview on these slides
- Current projects:
- Gamma rays classification and parameters estimation.
- Keywords : deep learning, astrophysics, model inversion, physics inspired models, domain adaptation, multitask learning, classification, regression.
- Academic partner: LAPP
- Industrial partner : Orobix
- Federated Learning and bias mitigation
- Keywords : deep learning, decentralized learning, model and data bias detection, clients clustering
- Academic and industrial research
- Deep learning for Alpin glacier model inversion
- Keywords : deep learning, model inversion, physics inspired models
- IoT and AI for health care
- Keywords : deep learning, time series analysis
- internal long run project at LISTIC
- Gamma rays classification and parameters estimation.
- Recent projects:
- Oil slicks detection and classification.
- Keywords : deep learning, SAR data analysis, heterogeneous data fusion, anomaly detection, semantic segmentation, classification.
- Industrial partner : Total
- Object detection and recognition for office inventory.
- Keywords : deep learning, 360° images, 3D scene modelling.
- Industrial partner : SmarterPlan.io
- Oil slicks detection and classification.
Teaching :
- Related to my research activities:
- Image and data processing with tools such as Matlab, and Python, OpenCV
- Deep larning with Tensorflow.
- Target students : PhD, Master and Bachelor degree and employed engineers.
- General teaching activities:
- Programming : scripting and object programming with C/C++/Java/Python/PHP, and others.
- Modeling (UML, etc.)
- Experience highlights :
- image processing and machine learning to PhD students and Master 2 at University Savoie Mont Blanc.
- machine learning at ESEN engineering school at Tunis (Tunisia)
- machine learning at Renault compagny (Paris, Versailles)
Consulting :
- Introducing machine learning in digital processes in the industry.
- Data pre-processing, features extraction and decision making for classification, regression and other tasks for image, video and time series understanding.
- Experience highlights :
- Machine Learning and Deep Learning at Renault (Paris, Versailles)
- Deep Learning at Total company(Paris)
- Machine Learning at AboutGoods Company and Inventhys (Annecy)
- Machine Learning with startups (Heliocity, Skiply and so on) along implications into the IDEFICS project
- AI project planning with Concentrix company
PhD thesis co-supervising :
Current PhDs:
- 2024-2028 François-Xavier Sikoumno
Topic : Decentralised federated deep learning, anomaly detection, multitemporal time series.
- 2022-2025 Lorenzo Lopez-Uroz
Topic : Deep learning, model inversion, physics inspired models.
- 2022-2025 Lynda Ferraguyg
Topic : Féderated learning, bias detection and mitigation
- 2021-2024 Michael Dell’Aiera
Topic : Semi supervised Deep neural networks for gamma rays analysis, physics inspired models..
Funding : H2020 ESCAPE project and LAPP lab collaboration.
Defended PhDs:
- 2018-2021 Emna Amri
Topic : Automatic Oil Slicks detection and classification from Remote Sensing imaging using Deep Learning.
Funding : Total company. - 2017-2020 Mickael Jacquemont
Topic : Deep neural networks for the 3D analysis of gamma rays.
Funding : H2020 ASTERICS project and the USMB fondation. - 2015-2019 Amina Ben Hamida
Topic : Indexing of large Remote sensing images.
Funding : French research ministry. - 2014-2019 Rizlène Raoui
Topic : Automatic Sales receipt understanding using Deep Learing.
Funding : CIFRE with AboutGoods Compagny. - 2013-2016 Nicolas Voiron
Topic : Automating clustering of multimedia data bases for exploration.
Funding : USMB - 2010-2013 Sabin Tiberius Strat
Topic : Automating understanding of image sequences using a collaborative approach
Funding : French research ministry
Conference/Workshop organization
- 2015 8th International Workshop on the Analysis of Multitemporal Remote Sensing Images, July 22-24, 2015 Annecy, France
- 2012 10th Workshop on Content-Based Multimedia Indexing (CBMI2012), 27-29 juin 2012, Annecy le Vieux, France.
- 2011 7th Conference of the European Society for Fuzzy Logic and Technology (EUSFLAT 2011) July 18-22, 2011, jointly with Les Rencontres Francophones sur la Logique Floue et ses Applications (LFA 2011, aka French Days on Fuzzy Logic and Applications) in Aix-les-Bains, France.
Reading committee
Journals:
IEEE Transactions on Image Processing ; Elsevier Computer Vision and Image Understanding ; Elsevier Signal Processing :Image Communication ; Multimedia Tools and APplications (MTAP) ; International Journal of Neural Systems ; MDPI Open Access Journal – Remote Sensing ; Hindawi – JournalOfRobotics ; ISPRS Journal of Photogrammetry and Remote Sensing ; JSTARS – GRSS | IEEE | Geoscience & Remote Sensing Society ; Integrated Computer-Aided Engineering
Conferences:
Main ones : IEEE International Conference on Image Processing ; International Workshop on Content-Based Multimedia Indexing ; JURSE – Joint Urban Remote Sensing Event ;
Scientific animation
2018 – 2019 co coordination of the MAIM at GDR ISIS.
2015 – 2017 co coordination of the IRIM at GDR ISIS..
Codes :
- a bio-inspired retina model for image pre-processing. contributed to the OpenCV library (C++/Python):
- Deep Learning frameworks and related tools :
- pyTorch based framework to conduct versioned experiments, codes&docs. This is developed and applied in the context of a phD focused on gamma rays detection and analysis in the context of the internationnal Cherenkov Telescope Array project. The proposed framework can be applied to any image processing problem.
- pyTorch indexed convolutions than enable any kind of kernels and pixel sampling grids : doc. This is applied to hexagonal images processing in the Cherenkov Telescope Array project.
- Tensorflow based framework to conduct versioned experiments, codes&docs. Several examples provided for image segmentation, curves regression, GAN, etc.
- caffe based deep learning model for hyperspectral images analysis using lightweight 3D convolutions based networks codes, paper