Information processing: Signals, Images, Graphs
Contacts for this research theme
Patrice Abry, Benjamin Audit, Pierre Borgnat, Patrick Flandrin, Nicolas Garnier, Pablo Jensen, Adrien Meynard, Nelly Pustelnik, Stéphane Roux, Julian Tachella
Inverse problems and optimisation
Inverse problems encompass the challenges of recovering missing information from limited or degraded measurements. In the recent years we have contributed to: revisiting of piece- wise constant or piecewise linear signal estimation ; image reconstruction, both from variational approaches to methods based on neural networks ; combining reconstruction with contour detection ; texture segmentation.
Self-Supervised Learning, Equivariance, and model-based NN for Inverse Problems
Improve the robustness and explainability of DL methods ; equi/invariance to transformations to stabilize DL ; model-based neural network architectures that incorporate knowledge about the acquisition physics.
Nonstationary signal processing
The group has a strong history of studying nonstationary signal processing, ranging from theoretical analalysis or development of approaches such as time-frequency methods, data-driven decompositions (e.g., Empirical Mode Decomposition), practical characterization of stationarity and non-stationarity. Many applications of these methods have been put forward as well by the group.
Multiscale analaysis for multivariate signals and fields: theory and applications
Modern applications often entails multivariate (e.g., neuros-ciences), heterogeneous (e.g., textures, heart rate variability) data defined on sparse supports (e.g., geolocalized geographical data). In the recent years, we have tackled these challenges and advanced developments on global multi-variate scale-free analysis.
Graph signals processing ; Complex networks
The team is active in the recent domain of Graph Signal Processing, as well as in various approaches for graph-based learning and analysis of complex networks.
Neuromorphic processing
Development of tools for neuromorphic processing, for event-based cameras, relying on inverse problems and nonstationary analysis