Céline Hudelot – CentraleSupelec
Open-Set Likelihood Maximization for Few-Shot Learning
Céline and her team tackle the Few-Shot Open-Set Recognition (FSOSR) problem, i.e. classifying instances among a set of classes for which we only have a few labeled samples, while simultaneously detecting instances that do not belong to any known class. We explore the popular transductive setting, which leverages the unlabelled query instances at inference. Motivated by the observation that existing transductive methods perform poorly in open-set scenarios, we propose a generalization of the maximum likelihood principle, in which latent scores down-weighing the influence of potential outliers are introduced alongside the usual parametric model. Our formulation embeds supervision constraints from the support set and additional penalties discouraging overconfident predictions on the query set. We proceed with a block-coordinate descent, with the latent scores and parametric model co-optimized alternately, thereby benefiting from each other. We call our resulting formulation Open-Set Likelihood Optimization (OSLO). OSLO is interpretable and fully modular; it can be applied on top of any pre-trained model seamlessly. Through extensive experiments, we show that our method surpasses existing inductive and transductive methods on both aspects of open-set recognition, namely inlier classification and outlier detection. Code is available at https://github.com/ebennequin/few-shot-open-set.
Alexis Joly – Inria, Montpellier University, LIRMM
Weakly-supervised analysis of multi-label plant images in the context of the Pl@ntNet project
This talk focuses on analyzing plant images in a multi-label context using weakly-supervised learning techniques. The aim is to propose methods that effectively handle partially or incompletely annotated data while maximizing the performance of classification models. Pl@ntNet, a platform for plant recognition based on images, provides the framework to illustrate these approaches and their applications in the fields of botany and ecology.
Nicolas Courty – Université Bretagne Sud, IRISA
Dimensionality reduction and Clustering through the lens of Optimal Transport
In this talk Nicolas will discuss how optimal transport can serve as a building block for dimensionality reduction (DR) and clustering tasks, generalizing well known concepts from data analysis. He will first present the SNEkhorn method [1], where he and his team uncover a novel characterization of entropic affinities, used in DR, as an optimal transport problem, and he will cover the idea of Gromov-Wasserstein projections as a general framework for joint-clustering and DR [2].
[1] https://arxiv.org/abs/2305.13797 (NeurIPS 2023)
[2] https://arxiv.org/abs/2402.02239 (Under submission)
Ronan Fablet – IMT Atlantique, Lab-STICC
Bridging Physics and Deep Learning for Ocean Modeling and Monitoring: How to deal with sparsely-sampled data?
Ronan is a Professor at IMT Atlantique and a research scientist at Lab-STICC, specializing in Data Science and Computational Imaging. His research focuses on the intersection of data science and ocean science, particularly in space oceanography and marine ecology. His work includes using deep learning for dynamical systems to analyze, simulate, and reconstruct ocean dynamics from satellite ocean remote sensing data. In this talk, « Bridging Physics and Deep Learning for Ocean Modeling and Monitoring: How to Deal with Sparsely-Sampled Data? » Ronan will address challenges in using deep learning to model and monitor the ocean, particularly in scenarios where data is limited. This topic explores how combining physics-based models with data-driven approaches can enhance our understanding of complex ocean processes, even when data is sparse.