Keywords
Team
STIM
Biography
Samy Blusseau is a researcher whose work lies at the intersection of image analysis, deep learning, and mathematical morphology. His research focuses primarily on the development of morphological neural networks, combining theoretical tools from mathematical morphology with deep learning architectures to solve complex problems in image processing. His contributions include studying the limitations of gradient-based optimization methods for morphological networks, as well as developing neural pipelines that incorporate geodesic reconstruction layers for tasks such as object counting in microscopic imaging. His recent work also explores the application of these methods to various fields, such as the modeling of composite materials, the mechanics of polycrystalline microstructures, and medical image segmentation, with an emphasis on the interpretability and efficiency of the proposed models.
Publication(s)
-
2026
Learning Morphological Representations of Image Transformations: Influence of Initialization and Layer Differentiability DOI : 10.1007/978-3-032-09544-2_27
-
2026
Optimizing Morphological Representations: Robustness to Initialization and Gradient Sparsity DOI : 10.1007/s10851-026-01331-8
-
2025
Cell Counting with Trainable h-Maxima and Connected Component Layers DOI : 10.1007/s10851-025-01243-z
-
2025
A physics-informed 3D surrogate model for elastic fields in polycrystals DOI : 10.1016/j.cma.2025.117944
-
2025
Two-Dimensional Stochastic Structural Geomodeling with Deep Generative Adversarial Networks DOI : 10.1007/s11004-025-10188-3
-
2024
Automatic yarn path extraction of large 3D interlock woven fabrics with confidence estimation DOI : 10.1016/j.compositesa.2024.108396
-
2024
Counting Melanocytes with Trainable h-Maxima and Connected Component Layers DOI : 10.1007/978-3-031-57793-2_32
-
2024
Training Morphological Neural Networks with Gradient Descent: Some Theoretical Insights DOI : 10.1007/978-3-031-57793-2_18
-
2023
Moving Frame Net: SE(3)-Equivariant Network for Volumes DOI : 10.48550/arXiv.2211.03420
-
2022
Scale-Equivariant U-Net DOI : 10.48550/arXiv.2210.04508
-
2022
Quantitative Characterization of Ductility for Fractographic Analysis DOI : 10.1007/978-3-031-11818-0_46
-
2022
Instance segmentation of 3D woven fabric from tomography images by Deep Learning and morphological pseudo-labeling DOI : 10.1016/j.compositesb.2022.110333
-
2022
APPLYING DEEP LEARNING TO MELANOCYTE COUNTING ON FLUORESCENT TRP1 LABELLED IMAGES OF IN VITRO SKIN MODEL DOI : 10.5566/ias.2640
-
2022
Morphological Adjunctions Represented by Matrices in Max-Plus Algebra for Signal and Image Processing DOI : 10.1007/978-3-031-19897-7_17
-
2022
DIFFERENTIAL INVARIANTS FOR SE(2)-EQUIVARIANT NETWORKS DOI : 10.1109/ICIP46576.2022.9897301
-
2022
Adaptive Anisotropic Morphological Filtering Based on Co-Circularity of Local Orientations DOI : 10.5201/ipol.2022.397
-
2021
On Some Associations Between Mathematical Morphology and Artificial Intelligence DOI : 10.1007/978-3-030-76657-3_33
-
2021
Scale Equivariant Neural Networks with Morphological Scale-Spaces DOI : 10.1007/978-3-030-76657-3_35
-
2019
Part-based approximations for morphological operators using asymmetric auto-encoders DOI : 10.1007/978-3-030-20867-7_25
-
2019
Max-plus operators applied to filter selection and model pruning in neural networks DOI : 10.1007/978-3-030-20867-7_24
-
2018
Tropical and Morphological Operators for Signals on Graphs DOI : 10.1109/ICIP.2018.8451395
-
2016
Measuring the visual salience of alignments by their non-accidentalness DOI : 10.1016/j.visres.2015.08.014
-
2014
A psychophysical evaluation of the a contrario detection theory DOI : 10.1109/ICIP.2014.7025217
Teaching
Differential, Integral, and Stochastic Calculus I (Math1)
This first module of UE11 will begin with an introduction to the topology of metric spaces: open sets, closed sets, density, completeness, compactness, and the space of continuous maps between metric spaces. We will then cover the essential elements of measure theory (without doing all the proofs in lecture), which allows us to define the Lebesgue integral; we will clarify how it differs from the Riemann integral covered in preparatory classes. This integral will allow us to define the main functional spaces used in the mathematical study of equations in physics—the Lp spaces—and in particular the L2 space. We will then study Hilbert spaces (of which L2 is an archetypal example), which are generalizations of Euclidean spaces to infinite dimensions, and we will examine their main properties, in particular the existence of so-called Hilbert bases. The final part of the course will be devoted to differential calculus for mappings from R^n to R^m: the concept of a partial derivative, the differential of a mapping, the finite increase theorem in R^d, the implicit function theorem, and the local inversion theorem. Advanced tutorial sessions will allow motivated students to go beyond the concepts required for the exam: more in-depth analysis of measure and integration theory, functional analysis, Banach spaces, the Hahn-Banach theorem, operators in infinite-dimensional spaces, Sobolev spaces… This theoretical course aims to provide the foundations that will enable students to tackle the major challenges of applied mathematics, in particular the study and numerical solution of differential equations, partial differential equations, and optimization.
Differential, Integral, and Stochastic Calculus II (Math2)
EC2 consists of two modules: Differential Equations and Probability. The differential equations course aims to introduce students to the study of dynamical systems: existence, uniqueness, and regularity of solutions to a differential equation (Peano–Arzela, Cauchy-Lipschitz, regularity with respect to initial conditions in finite time, chaotic systems), as well as an introduction to the study of the asymptotic behavior of solutions (periodic cycles, asymptotic stability of equilibrium points, etc.) An introduction to the simulation and numerical analysis of differential equations is also provided (discretization schemes, consistency/convergence analysis, differences between explicit and implicit schemes for stiff systems, the role of symplectic schemes for Hamiltonian systems, etc.) The probability course aims to consolidate and supplement the knowledge of probability theory acquired in CPGE, but above all to develop probabilistic reasoning. In CPGE, probability was studied in the context of random phenomena with at most a countable number of possible outcomes. Probabilities defined on the real line, as well as real random variables and vectors, are first introduced within the general formalism of measure theory—covered in EC1—which allows for the inclusion of the discrete case. The concepts of independence and conditioning of random variables, sequences of random variables, and finally stochastic simulation methods are addressed in turn to cover all the prerequisites necessary for the various engineering specializations offered at the school, particularly data science.
Signal Processing
The Signal Processing course covers all aspects of harmonic analysis, including discrete- and continuous-time signals, convolution, operational calculus, and the Fourier transform. This content is presented in a way that makes it applicable in various contexts such as mathematics, engineering, and physics. The course establishes a fundamental link between the mathematical foundations of signal processing (Fourier analysis, wavelets, etc.) and the practical tools derived from them, such as filtering and compression. These concepts are illustrated by modern technologies that incorporate these principles, thereby providing a concrete and applied perspective on the theories covered. Furthermore, the course highlights the strong interconnections between signal processing and other fields covered by the Applied Mathematics course unit, naturally guiding instruction toward the fundamental mathematical concepts of the field. Practical exercises, designed to simply illustrate the theorems studied in class, reinforce this theoretical learning. Finally, the exploration of recent technological applications provides a current and dynamic perspective on the theoretical results presented. Independent study hours are devoted to projects on topics that go beyond the scope of the course. Recent projects have included the implementation of a music recognition algorithm based on the windowed Fourier transform, the study of signal processing tools used in tomography, and the application of signal processing tools to a geophysical problem.
PhD supervision
- 2025 Robustness by design for frugal and trustworthy learning models LOZA RAMIREZ Edgar
- 2024 Hybrid image processing for surface particulate contamination metrology: coupling morphological methods and learning-based methods for aerosols PILLARD Dorian
- 2024 Morphological layers in neural networks: how to train and use them for data analysis DIMITROVA Mihaela
- 2022 Stochastic two-dimensional structural geomodeling using deep generative adversarial networks GARAYT Charlie
- 2022 Estimation of mechanical properties of polycrystals through physics-informed machine learning MONTEIRO FERNANDES Lucas
- 2019 Deep Learning Equivariant Based on Scale Spaces and Moving Frames SANGALLI Mateus