Seminars – 2024-2025

  • Date: August, 26th
  • Room: Escale en Arz, in Arz island.
  • Title: Seminaire au vert! (Annual team seminar)
  • Program
  • Date: Wednesday, October 23rd – 10:00
  • Room: A103, ENSIBS Vannes
  • Speaker: Tom Avellaneda (IRISA Obelix) – Internship oral presentation
  • Title: Efficient Remote Sensing Change Retrieval without Learning using Pattern Spectra
  • Abstract: In the context of remote sensing image analysis and earth observation, this study focuses on the problem of spatio-temporal patch retrieval. Given a small pattern with spatial and temporal characteristics, we want to retrieve similar patterns from a very large satellite image time series. To solve this problem we leverage pattern spectra, attribute histograms which can serve as features and can be computed very efficiently using morphological hierarchies. The extension of pattern spectra to spatio-temporal data has not yet been fully explored, and their potential use for spatio-temporal pattern retrieval has not been tested. We propose a large variety of possible patch retrieval models using pattern spectra, which can be computed as a grid very efficiently, and test the performance of these approaches on the SECOND change detection dataset.
  • Date: WEDNESDAY, November 13th – afternoon
  • Room : in Rouen
  • Speaker: Guillaume MAHEY (IRISA Obelix) — Ph.D. defense
  • Title : Unbalanced and Linear Optimal Transport for Reliable Estimation of the Waserstein Distance
  • Abstract: Despite being very elegant in theory, optimal transport (OT) might suffer from several drawbacks in practice. Notably, the computational burden, the risk of over fitting and its sensibility to artifacts motivated the introduction of variants for the OT loss in the ML community. In this thesis, we propose such variants in order to, first, reduce computational and statistical burden and second the sensibility to artifacts of sampling of the OT loss. To do that, we built on the top of proxies distributions introduced both by the Linear and the Unbalanced OT variants.spaces for representation learning by proposing a new classification method in hyperbolic spaces. Our method which is hierarchically-informed improves the performance in hierarchical classification.

as Guillaume’s defense will be held in Rouen, he will repeat his presentation in Vannes on Friday, 13th december at 11:00 am / Room A102 (ENSIBS building)

  • Date: Thursday, November 28th – 11:30
  • Room: A102, ENSIBS Vannes
  • Speaker: Titouan Vayer (researcher Inria–OCKHAM, ENS Lyon)
  • Title: Learning graphs with precision matrices: statistical estimators, compressive approches and unrolled neural networks
  • Abstract: Learning a graph of relationships between variables from signals is a fundamental problem, particularly for analyzing geographic data or EEG brain data. Such a graph provides insights into the statistical dependencies between key variables. In this presentation, I will first review standard statistical approaches and convex algorithms or solving this problem, then introduce two distinct works: one focused on inverse problems, and the other on data-driven neural network approaches. The first work demonstrates how to estimate such a graph from a sketch of the data, which is a small summary built from random projections. I will show that, from an information-theoretic perspective, it is possible to recover the graph from a limited number of measurements, under sparsity assumptions. In the second part, I will present a novel neural network architecture, called SpodNet (Schur’s Positive-Definite Network), which tackles the graph estimation problem in a data-driven manner.
  • Date: FRIDAY, December  13th – afternoon
  • Room : Amphi ENSIBS, Vannes
  • Speaker: Paul BERG (IRISA Obelix) — Ph.D. defense
  • Title : Contributions to Representation Learning in Computer Vision and Remote Sensing
  • Abstract : Deep Learning has become an ubiquitous tool for the resolution of image analysis tasks, notably in the remote sensing domain. As such, the need for annotated data have largely increased. But, annotating data can be costly and time consuming. Therefore, a whole field of the literature is dedicated to image representation learning by decreasing the dependence to annotations using so called self-supervised methods. The learnt representations are then usable in downstream tasks because of their discriminative nature with respect to the labels. In this context, we evaluate in this thesis how these methods can be exploited in the remote sensing domain by investigating tasks such as multi-modal scene classification for which we propose a self-supervised framework. We leverage the optimal transport problem to model several problems et propose new methodological contributions to contrastive learning. Finally, we propose to go beyond Euclidean spaces for representation learning by proposing a new classification method in hyperbolic spaces. Our method which is hierarchically-informed improves the performance in hierarchical classification.
  • Date: Thursday, February 27, 2025 at 11:30 a.m.
  • Room: A102, ENSIBS Vannes
  • Speaker: Antoine Bralet
  • Title: Deep learning for multimodal detection of sudden and slow moving slope instabilities on bitemporal remote sensing images
  • Abstract: The aim of this work is to develop artificial intelligence algorithms for the detection of sudden and slow slope instabilities from synthetic aperture radar (SAR) and optical satellite images. The first contributions led to the creation of a new neural network architecture, SARDINet, enabling SAR images to be translated into optical images, so that comparable images can be obtained independently of weather conditions. Several architectural modifications improved the relevance and contrast of the translated images. These translations were then conditioned using new cost functions to guide the translator in delivering images that can be used a posteriori for classification or change detection tasks. These translations are then applied to the detection of sudden landslides in Haiti in 2021. Finally, the detection of slow instabilities was the subject of the creation of a new dataset, ISSLIDE. The dataset is based on a new manual routine of annotation on SAR interferograms and exploited by standard segmentation neural networks. It is also used to develop a new segmentation strategy named ECSPLAIN. The latter relies on the explainability of a classifier network during the training phase to generate segmentation maps and force the classifier to be explicable. Thus, this work provides new artificial intelligence methods applied to satellite images, while meeting the need for understanding in the geosciences.
  • Date: Thursday, March 6, 2025 at 11:30 a.m.
  • Room: A102, ENSIBS Vannes
  • Speaker: Ambroise Odonnat
  • Title: MaNo: Exploiting Matrix Norm for Unsupervised Accuracy Estimation under Distribution Shifts
  • Abstract: Leveraging the model’s outputs, specifically the logits, is a common approach to estimating the test accuracy of a pre-trained neural network on out-of-distribution (OOD) samples without requiring access to the corresponding ground-truth labels. Despite their ease of implementation and computational efficiency, current logit-based methods are vulnerable to overconfidence issues, leading to prediction bias, especially under the natural shift. In this work, we first study the relationship between logits and generalization performance from the view of low-density separation assumption. Our findings motivate our proposed method MANO that (1) applies a data-dependent normalization on the logits to reduce prediction bias, and (2) takes the Lp norm of the matrix of normalized logits as the estimation score. Our theoretical analysis highlights the connection between the provided score and the model’s uncertainty. We conduct an extensive empirical study on common unsupervised accuracy estimation benchmarks and demonstrate that MANO achieves state-of-the-art performance across various architectures in the presence of synthetic, natural, or subpopulation shifts.
  • Date: Thursday, March 20, 2025 at 10:30 a.m.
  • Room: A102, ENSIBS Vannes
  • Speaker: Mareike Dorozynski
  • Title: Multi-temporal multi-modal classification of geodata for spatiotemporal analysis of the landscape
  • Abstract: Geoinformation, especially information on land cover, is not only of interest for describing the current state of the Earth’s surface but also for past, historical states. Land cover time series can be utilized to analyse the development and change of the environment, to identify various spatial processes – such as changes in cities, forests and water bodies –, and to gain an understanding of interdependencies between processes. Therefore, such knowledge is essential to make well-founded decisions about the design of future habitats. Against this background, it is highly relevant to develop methods for the automated classification of geodata of different ages and quality to enable computer-aided analysis of time series. Available data sources – of multiple modalities, scales, and epochs – should be combined so that a classifier can benefit from the strengths of all considered data. Examples of multi-modal, multi-temporal, and multi-temporal multi-scale classification approaches will be presented. Furthermore, an outlook on the next steps in multi-temporal multi-modal land cover classification will be provided.
  • Date: Thursday, April 3, 2025 at 11 a.m.
  • Room: A104, ENSIBS Vannes
  • Speaker: Thibault de Surrel
  • Title: Wrapped Gaussian on the manifold of Symmetric Positive Definite matrices
  • Abstract: Circular and non-flat data distributions are prevalent across diverse domains of data science, yet their specific geometric structures often remain underutilized in machine learning frameworks. A principled approach to accounting for the underlying geometry of such data is pivotal, particularly when extending statistical models, like the pervasive Gaussian distribution. In this presentation, I will tackle those issue by focusing on the manifold of symmetric positive definite matrices, a key focus in information geometry. I will introduced a non-isotropic wrapped Gaussian by leveraging the exponential map, derive theoretical properties of this distribution and propose a maximum likelihood framework for parameter estimation. Furthermore, I will reinterpret established classifiers on SPD through a probabilistic lens and introduce new classifiers based on the wrapped Gaussian model. This work lays the groundwork for extending classical machine learning and statistical methods to more complex and structured data.
  • Date: Thursday, April 24, 2025 at 11:30 a.m.
  • Room: A102, ENSIBS Vannes
  • Speaker: Lucie Laporte-Devylder
  • Title: Seeing the Unseen: A Thermal Perspective on Marine Mammal Monitoring
  • Abstract: In this seminar, I will present ongoing work as part of WildDrone, a multidisciplinary EU MSCA project leveraging drone technology to address biodiversity challenges. While the project develops advanced machine learning tools for automation, my focus as a biologist is on translating these innovations into field-ready methods for marine mammal monitoring. Traditional monitoring techniques, such as GPS tagging, can be invasive and limited in scope. Advances in drone technology and infrared (IR) imaging now offer promising, non-invasive alternatives. IR enables both direct detection of marine mammals via thermal signatures, and indirect detection through flukeprints and other water disturbances. Yet, the underlying mechanisms influencing these thermal cues, their environmental dependencies, and their utility for species identification, age classification, and behavioral tracking remain largely unexplored. In 2024, I conducted drone-based field trials in La Réunion during the humpback whale (Megaptera novaeangliae) breeding season, using thermal and RGB cameras to assess the reliability of flukeprints for detection and tracking. My next campaign in May 2025 will expand this work to Arctic environments, exploring how thermal cue characteristics shift across climates. Looking forward, this research opens key opportunities for computer vision applications. These include automated detection of thermal tracks, diffusion-based modeling to estimate whale time-of-passage, and real-time species or individual identification using visual features from drone imagery. By bridging marine biology, drone sensing, and machine learning, this work aims to unlock new tools for non-invasive marine mammal monitoring and catalyze interdisciplinary collaboration in conservation science.
  • Date: Tuesday, June 10, 2025 at 11 a.m.
  • Room: A102, ENSIBS Vannes
  • Speaker: Rostislav Netek. Dr. Rostislav Netek is an Assistant Professor at Dept. of Geoinformatics , Palacký University Olomouc. His professional skills and interests are Geoinformatics, especially he is focused on Web cartography, WebGIS solutions and Open source technologies. He is an author of the first book about a Web cartography. He is a member of ICA Commission on Maps and the Internet, member of Expert group in Czech Association of Geoinformatics and member of OSGeo.
  • Title: Map tiles – (r)evolution in web cartography
  • Abstract: In 2005 Google Maps launched “slippy maps” – a brand new concept of map tiles how to handle with spatial data for web maps. Nowadays, map tiles are routinely used by all major web map solutions such as Google, Esri, Bing, OpenStreetMap or Mapbox as a standard. This presentation explores the evolution, current state, and and both raster and vector approaches of map tiling technologies, arguing that they represent a technical revolution as well as a cartographic paradigm shift in how geospatial information is produced, distributed, and consumed online.
  • Date: Thursday, June 12, 2025 at 11:30 a.m.
  • Room: A102, ENSIBS Vannes
  • Speaker: Anne Gagneux
  • Title: Plug-and-Play methods: theory and practice
  • Abstract: In image restoration, PnP methods leverage the strength of trainable denoisers by integrating them in existing optimization schemes. First, we will show how to leverage generative models to create new denoisers. Specifically, we introduce the PnPFlow algorithm, a PnP method based on Flow Matching. In the second part of the talk, we will study desirable properties of PnP denoisers that ensure convergence of the associated iterative schemes. In particular, we provide an in-depth study of necessary and sufficient conditions for a neural network to be convex, beyond the traditional Input Convex Neural Network (ICNN) architecture.
  • Date: Thursday, July 10, 2025 at 11:30 a.m.
  • Room: A102
  • Speaker: David Coudert
  • Title: How to compute the hyperbolicity of large graphs
  • Abstract: The notion of hyperbolicity in graphs is a measure of how much a graph resembles a tree from a metric perspective. This notion has been used in different contexts, such as the design of routing schemes, network security, computational biology, the analysis of graph algorithms, and the classification of complex networks. For instance, it gives bounds on the best possible stretch of some greedy-routing schemes in Internet-like graphs. Its computation is challenging as the main approaches consist in scanning all quadruples of the graph or using fast matrix multiplication as building block. Both approaches are not practical for large graphs. The best known algorithm has time complexity in O(n^{3.69}) and space complexity in O(n^2). In this talk, we will survey the different methods that have been proposed to enable the computation of the hyperbolicity of graphs with up to a million nodes.