Céline Hudelot

CentraleSupélec | ECP · Laboratory of Applied Mathematics and Systems (MAS) - EA 4037

Class-Incremental Learning (CIL) aims to build classification models from data streams. At each step of the CIL process, new classes must be integrated into the model. Due to catastrophic forgetting, CIL is particularly challenging when examples from past classes cannot be stored, the case on which we focus here. To date, most approaches are based...

We 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. Motivat...

Prediction of the next job application is one of the most important use-cases of job recommender systems. This work proposes to use next-item recommendation methods to model job seekers’ career preferences to more accurately discover the next job postings they may apply for. Our proposed model, Personalized-Attention Next-Application Prediction (PA...

Machine Learning models are prone to fail when test data are different from training data, a situation often encountered in real applications known as distribution shift. While still valid, the training-time knowledge becomes less effective, requiring a test-time adaptation to maintain high performance. Following approaches that assume batch-norm l...

We tackle the Few-Shot Open-Set Recognition (FSOSR) problem, i.e. classifying instances among a set of classes for which we only have few labeled samples, while simultaneously detecting instances that do not belong to any known class. Departing from existing literature, we focus on developing model-agnostic inference methods that can be plugged int...

Deep learning approaches are successful in a wide range of AI problems and in particular for visual recognition tasks. However, there are still open problems among which is the capacity to handle streams of visual information and the management of class imbalance in datasets. Existing research approaches these two problems separately while they co-...

Every day, a new method is published to tackle Few-Shot Image Classification, showing better and better performances on academic benchmarks. Nevertheless, we observe that these current benchmarks do not accurately represent the real industrial use cases that we encountered. In this work, through both qualitative and quantitative studies, we expose...

Works on learning job title representation are mainly based on \textit{Job-Transition Graph}, built from the working history of talents. However, since these records are usually messy, this graph is very sparse, which affects the quality of the learned representation and hinders further analysis. To address this specific issue, we propose to enrich...

With the increase in use of machine learning classifiers in several fields, providing human- understandable explanation of their outputs has become an imperative. It is essential to generate trust for day-to-day tasks, especially in the sensible domains as medical imaging. Although many works have addressed this problem by generating visual explana...

Active learning aims to optimize the dataset annotation process when resources are constrained. Most existing methods are designed for balanced datasets. Their practical applicability is limited by the fact that a majority of real-life datasets are actually imbalanced. Here, we introduce a new active learning method which is designed for imbalanced...

Deep learning approaches are successful in a wide range of AI problems and in particular for visual recognition tasks. However, there are still open problems among which is the capacity to handle streams of visual information and the management of class imbalance in datasets. Existing research approaches these two problems separately while they co-...

When we can not assume a large amount of annotated data , active learning is a good strategy. It consists in learning a model on a small amount of annotated data (annotation budget) and in choosing the best set of points to annotate in order to improve the previous model and gain in generalization. In deep learning, active learning is usually imple...

This paper tackles the problem of processing and combining efficiently arbitrary long data streams, coming from different modalities with different acquisition frequencies. Common applications can be, for instance, long-time industrial or real-life systems monitoring from multimodal heterogeneous data (sensor data, monitoring report, images, etc.)....

In this paper we explore contrastive learning for few-shot classification, in which we propose to use it as an additional auxiliary training objective acting as a data-dependent regularizer to promote more general and transferable features. In particular, we present a novel attention-based spatial contrastive objective to learn locally discriminati...

Few-Shot Learning (FSL) algorithms have made substantial progress in learning novel concepts with just a handful of labelled data. To classify query instances from novel classes encountered at test-time, they only require a support set composed of a few labelled samples. FSL benchmarks commonly assume that those queries come from the same distribut...

Explaining the decisions of deep learning models is critical for their adoption in medical practice. In this work, we propose to unify existing adversarial explanation methods and path-based feature importance attribution approaches. We consider a path between the input image and a generated adversary and associate a weight depending on the model o...

Domain Invariant Representations (IR) has improved drastically the transferability of representations from a labelled source domain to a new and unlabelled target domain. Unsupervised Domain Adaptation (UDA) in presence of label shift remains an open problem. To this purpose, we present a bound of the target risk which incorporates both weights and...

Providing a human-understandable explanation of classifiers' decisions has become imperative to generate trust in their use for day-to-day tasks. Although many works have addressed this problem by generating visual explanation maps, they often provide noisy and inaccurate results forcing the use of heuristic regularization unrelated to the classifi...

Few-Shot Learning (FSL) algorithms have made substantial progress in learning novel concepts with just a handful of labelled data. To classify query instances from novel classes encountered at test-time, they only require a support set composed of a few labelled samples. FSL benchmarks commonly assume that those queries come from the same distribut...

With the recent successes of black-box models in Artificial Intelligence (AI) and the growing interactions between humans and AIs, explainability issues have risen. In this article, in the context of high-stake applications, we propose an approach for explainable classification and annotation of images. It is based on a transparent model, whose rea...

Unsupervised Domain Adaptation (UDA) has attracted a lot of attention in the last ten years. The emergence of Domain Invariant Representations (IR) has improved drastically the transferability of representations from a labelled source domain to a new and unlabelled target domain. However, a potential pitfall of this approach, namely the presence of...

Existing few-shot classification methods rely to some degree on the cross-entropy (CE) loss to learn transferable representations that facilitate the test time adaptation to unseen classes with limited data. However, the CE loss has several shortcomings, e.g., inducing representations with excessive discrimination towards seen classes, which reduce...

Among the wide variety of image generative models, two models stand out: Variational Auto Encoders (VAE) and Generative Adversarial Networks (GAN). GANs can produce realistic images, but they suffer from mode collapse and do not provide simple ways to get the latent representation of an image. On the other hand, VAEs do not have these problems, but...

Explaining decisions of black-box classifiers is paramount in sensitive domains such as medical imaging since clinicians confidence is necessary for adoption. Various explanation approaches have been proposed, among which perturbation based approaches are very promising. Within this class of methods, we leverage a learning framework to produce our...

Unsupervised Domain Adaptation (UDA) aims to bridge the gap between a source domain, where labelled data are available, and a target domain only represented with unlabelled data. If domain invariant representations have dramatically improved the adaptability of models, to guarantee their good transferability remains a challenging problem. This pape...

In this work, we propose a new unsupervised image segmentation approach based on mutual information maximization between different constructed views of the inputs. Taking inspiration from autoregressive generative models that predict the current pixel from past pixels in a raster-scan ordering created with masked convolutions, we propose to use dif...

In this work, we propose a new unsupervised image segmentation approach based on mutual information maximization between different constructed views of the inputs. Taking inspiration from autoregressive generative models that predict the current pixel from past pixels in a raster-scan ordering created with masked convolutions, we propose to use dif...

Learning Invariant Representations has been successfully applied for reconciling a source and a target domain for Unsupervised Domain Adaptation. By investigating the robustness of such methods under the prism of the cluster assumption, we bring new evidence that invariance with a low source risk does not guarantee a well-performing target classifi...

Unsupervised Domain Adaptation (UDA) has attracted a lot of attention in the last ten years. The emergence of Domain Invariant Representations (IR) has improved drastically the transferability of representations from a labelled source domain to a new and unlabelled target domain. However, a potential pitfall of this approach, namely the presence of...

Deep neural networks demonstrated their ability to provide remarkable performances on particular supervised learning tasks (e.g., image classification) when trained on extensive collections of labeled data (e.g., ImageNet). However, creating such large datasets requires a considerable amount of resources, time, and effort. Such resources may not be...

In this paper, we present a novel cross-consistency based semi-supervised approach for semantic segmentation. Consistency training has proven to be a powerful semi-supervised learning framework for leveraging unlabeled data under the cluster assumption, in which the decision boundary should lie in low-density regions. In this work, we first observe...

Recent deep generative models are able to provide photo-realistic images as well as visual or textual content embeddings useful to address various tasks of computer vision and natural language processing. Their usefulness is nevertheless often limited by the lack of control over the generative process or the poor understanding of the learned repres...

Despite the recent successes of deep learning, such models are still far from some human abilities like learning from few examples, reasoning and explaining decisions. In this paper, we focus on organ annotation in medical images and we introduce a reasoning framework that is based on learning fuzzy relations on a small dataset for generating expla...

A representation is supposed universal if it encodes any element of the visual world (e.g., objects, scenes) in any configuration (e.g., scale, context). While not expecting pure universal representations, the goal in the literature is to improve the universality level, starting from a representation with a certain level. To improve that universali...

Many real-world visual recognition use-cases can not directly benefit from state-of-the-art CNN-based approaches because of the lack of many annotated data. The usual approach to deal with this is to transfer a representation pre-learned on a large annotated source-task onto a target-task of interest. This raises the question of how well the origin...

Many real-world visual recognition use-cases can not directly benefit from state-of-the-art CNN-based approaches because of the lack of many annotated data. The usual approach to deal with this is to transfer a representation pre-learned on a large annotated source-task onto a target-task of interest. This raises the question of how well the origin...

This document reports some supplementary material that is not required to understand the main article but provide complements or illustrations. Hence, the additional elements were produced using the same version of the approach explained in our main paper and include the following items: (i) the detailed characteristics of the datasets used in this...

Association rules allow to mine large datasets to automatically discover relations between variables. In order to take into account both qualitative and quantitative variables, fuzzy logic has been applied and many association rule extraction algorithms have been fuzzified. In this paper, we propose a fuzzy adaptation of the well-known Close algori...

Capturing and understanding crowd dynamics is an important issue under diverse perspectives. From social, psychological, and political sciences to safety management, studying, modeling, and predicting the presence, behavior, and dynamics of crowds, possibly preventing dangerous activities, is absolutely crucial. In the literature, crowds have been...

Belief revision of knowledge bases represented by a set of sentences in a given logic has been extensively studied but for specific logics, mainly propositional, and also recently Horn and description logics. Here, we propose to generalize this operation from a model-theoretic point of view, by defining revision in the abstract model theory of sati...

In a transfer-learning scheme, the intermediate layers of a pre-trained CNN are employed as universal image representation to tackle many visual classification problems. The current trend to generate such representation is to learn a CNN on a large set of images labeled among the most specific categories. Such processes ignore potential relations b...

This paper tackles two recent promising issues in the field of computer vision, namely “the integration of linguistic and visual information” and “the use of semantic features to represent the image content”. Semantic features represent images according to some visual concepts that are detected into the image by a set of base classifiers. Recent wo...

Les représentations sémantiques décrivant les images par un ensemble de classifieurs de concepts ont montré des performances intéressantes en vision. Habituellement, toutes les sorties des classifieurs sont exploitées, mais il a été récemment montré que forcer le descrip-teur à être parcimonieux améliore les performances et la mise à l'échelle. Cep...

In this paper we extend some previously established links between the derivation operators used in formal concept analysis and some mathematical morphology operators to fuzzy concept analysis. We also propose to use mathematical morphology to navigate in a fuzzy concept lattice and perform operations on it. Links with other lattice-based formalisms...

We consider the problem of image classification with semantic features that are built from a set of base classifier outputs, each reflecting visual concepts. However, existing approaches consider visual concepts independently from each other whereas they are often linked together. When those relations are considered, existing models strongly rely o...

La classification d'images au moyen de descripteurs sé-mantiques repose sur des caractéristiques formées par les sorties de classifieurs binaires, chacun détectant un concept visuel dans l'image. Les approches existantes considèrent souvent les concepts visuels indépendam-ment les uns des autres, alors qu'ils sont souvent liés. Ces relations sont p...

We propose a method to complete description logic (DL) knowledge bases. For this, we firstly build a canonical finite model from a given DL knowledge base satisfying some constraints on the form of its axioms. Then, we build a new DL knowledge base that infers all the properties of the canonical model. This latter DL knowledge base necessarily comp...

As ontologies and description logics (DLs) reach out to a broader audience, several reasoning services are developed in this context. Belief revision is one of them, of prime importance when knowledge is prone to change and inconsistency. In this paper we address both the generalization of the well-known AGM postulates, and the definition of concre...

3D meshes are commonly used to represent virtual surface and volumes. However, their raw data representations take a large amount of space. Hence, 3D mesh compression has been an active research topic since the mid 1990s. In 2005, two very good review articles describing the pioneering works were published. Yet, new technologies have emerged since...

Belief revision of knowledge bases represented by a set of sentences in a given logic has been extensively studied but for specific logics, mainly propositional, but also recently Horn and description logics. Here, we propose to generalize this operation from a model-theoretic point of view, by defining revision in a categorical abstract model theo...

The paper proposes an ontology alignment framework with two core features: the use of background knowledge and the ability to handle imprecision in the matching process and the resulting concept alignments. The procedure is based on the use of a generic reference vocabulary, which is used to define an explicit semantic space for the ontologies to b...

In image interpretation and computer vision, spatial relations between objects and spatial reasoning are of prime importance for recognition and interpretation tasks. Quantitative representations of spatial knowledge have been proposed in the literature. In the Artificial Intelligence community, logical formalisms such as ontologies have also been...

In this paper, we propose an original way of enriching description logics with abduction reasoning services. Under the aegis of set and lattice theories, we put together ingredients from mathematical morphology, description logics, and formal concept analysis. We propose computing the best explanations of an observation through algebraic erosion ov...

We propose an ontology alignment framework with two core features: the use of background knowledge and the ability to handle vagueness in the matching process and the resulting concept alignments. The procedure is based on the use of a generic reference vocabulary, which is used for fuzzifying the ontologies to be matched. The choice of this vocabu...

This paper proposes a methodology for building fuzzy multimedia ontologies dedicated to image annotation. The built ontology incorporates visual, conceptual, contextual and spatial knowledge about image concepts in order to model image semantics in an e�ective way. Indeed, our approach uses visual and conceptual information to build a semantic hier...

This paper presents a new random accessible and progressive lossless manifold triangle mesh compression algorithm named POMAR. It allows to extract different parts of the input mesh at different levels of detail during the decompression. A smooth transition without artefacts is generated between adjacent regions decompressed at different levels of...

In this paper, we build upon previous work defining explanatory relations based on mathematical morphology operators on logical formulas in propositional logics. We propose to extend such relations to the case where the set of models of a formula is fuzzy, as a first step towards morphological fuzzy abduction. The membership degrees may represent d...

Although mathematical morphology and formal concept analysis are two lattice-based data analysis theories, they are still developed in two disconnected research communities. The aim of this paper is to contribute to fill this gap, beyond the classical relationship between the Galois connections defined by the derivation operators and the adjunction...

This paper proposes a new methodology to automatically build semantic hierarchies suitable for image annotation and classification. The building of the hierarchy is based on a new measure of semantic similarity. The proposed measure incorporates several sources of information : visual, conceptual and contextual as we defined in this paper. The aim is...

We address the problem of tag completion for automatic image annotation. Our method consists in two main steps: creating a list of “candidate tags” from the visual neighbors of the untagged image then using them as pieces of evidence to be combined to provide the final list of predicted tags. Both steps introduce a scheme to tackle with imprecision...

Annotating images using a fixed number of concepts is a fundamental task for content based image retrieval and classification. In practice, several modalities (visual, text..) provide information about the content of images. We are specifically interested in the tags associated with images, usually resulting from folksonomy, that provide imperfect...

We introduce the bag-of-multimedia-words model that tightly combines the heterogeneous information coming from the text and the pixel-based information of a multimedia document. The proposed multimedia feature generation process is generic for any multi-modality and aims at enriching a multimedia document description with compact and discriminative...

Semantic hierarchies have been introduced recently to improve image annotation. They was used as a framework for hierarchical image classification, and thus to improve classifiers accuracy and reduce the complexity of managing large scale data. In this paper, we investigate the contribution of semantic hierarchies for hierarchical image classificat...

This paper presents a new algorithm for the progressive compression of manifold polygon meshes. The input surface is decimated by several traversals that generate successive levels of detail through a specific patch decimation operator which combines vertex removal and local remeshing. The mesh connectivity is encoded by two lists of Boolean error...

The automatic attribution of semantic labels to unlabeled or weakly labeled images has received considerable atten-tion but, given the complexity of the problem, remains a hard research topic. Here we propose a unified classifica-tion framework which mixes textual and visual information in a seamless manner. Unlike most recent previous works, compu...

Classifier combination is known to generally perform better than each individ-ual classifier by taking into account the complementarity between the input pieces of information. Dempster-Shafer theory is a framework of interest to make such a fusion at the decision level, and allows in addition to handle the conflict that can exist between the class...