Publications

2018

• Taking Apart Autoencoders: How do They Encode Geometric Shapes ?
  
  • Newson Alasdair
  • Almansa Andrés
  • Gousseau Yann
  • Ladjal Saïd
  , 2018. We study the precise mechanisms which allow autoencoders to encode and decode a simple geometric shape, the disk. In this carefully controlled setting, we are able to describe the specific form of the optimal solution to the minimisation problem of the training step. We show that the autoencoder indeed approximates this solution during training. Secondly, we identify a clear failure in the generali-sation capacity of the autoencoder, namely its inability to interpolate data. Finally, we explore several regularisation schemes to resolve the generalisation problem. Given the great attention that has been recently given to the generative capacity of neural networks, we believe that studying in depth simple geometric cases sheds some light on the generation process and can provide a minimal requirement experimental setup for more complex architectures.
• Estimation d'un circuit électrique équivalent, à résistances et capacités thermiques, d'un bâtiment pour le contrôle optimal du chauffage du bâtiment
  
  • Nabil Tahar
  • Jicquel Jean-Marc
  • Girard Alexandre
  • Roueff François
  , 2018, pp.https://permalink.orbit.com/RenderStaticFirstPage?XPN=S5GmjW98%252BeXWqxLm4QBXD3fDUqlXTJ5uwQdFuycu4uk%3D%26n%3D1&id=0&base=FAMPAT. The invention relates to a method for determining a thermal model of a building equipped with a heating installation, in particular for energy diagnosis or optimization of the heating of said building, wherein: - An overall energy consumption load curve (CDC) is obtained from at least one energy consumption meter (COC), In predefined time steps, said load curve being capable of containing a consumption payload (Qu) for heating the building by said installation as well as a load for consumption needs not linked to the heating of the building, - And of one or more connected objects associated with respective appliances, which consume energy and are not controlled for a heat supply (ACN), at least one item of information on the switching on or off of said appliances (ACN), And time intervals are detected in the load curve (CDC) during which the connected objects inform of a stop state of the respective apparatuses, in order to obtain a first estimate of said payload (Qu), which makes it possible to iteratively correct the model for its optimization.
• Generalization Bounds for Minimum Volume Set Estimation based on Markovian Data, ISAIM, International Symposium on Artificial Intelligence and Mathematics proceedings, 1-7
  
  • Bertail Patrice
  • Ciołek Gabriela
  • Clémençon Stéphan
  , 2018.
• Profitable Bandits
  
  • Achab Mastane
  • Clémençon Stéphan
  • Garivier Aurélien
  Proceedings of Machine Learning Research, PMLR, 2018, 95, pp.694-709. Originally motivated by default risk management applications, this paper investigates a novel problem, referred to as the profitable bandit problem here. At each step, an agent chooses a subset of the K ≥ 1 possible actions. For each action chosen, she then respectively pays and receives the sum of a random number of costs and rewards. Her objective is to maximize her cumulated profit. We adapt and study three well-known strategies in this purpose, that were proved to be most efficient in other settings: kl-UCB, Bayes-UCB and Thompson Sampling. For each of them, we prove a finite time regret bound which, together with a lower bound we obtain as well, establishes asymptotic optimality in some cases. Our goal is also to compare these three strategies from a theoretical and empirical perspective both at the same time. We give simple, self-contained proofs that emphasize their similarities, as well as their differences. While both Bayesian strategies are automatically adapted to the geometry of information, the numerical experiments carried out show a slight advantage for Thompson Sampling in practice.
• Managing 'proto-ecosystems' - two smart mobility case studies
  
  • Marcocchia Giulia
  • Maniak Rémi
  International Journal of Automotive Technology and Management, Inderscience, 2018, 18 (3), pp.209-228. This paper considers how ecosystem-based research projects can be managed for a successful deployment of systemic and disruptive innovation. Such projects are defined as assignments in which heterogeneous organisations must invest upfront, aiming at co-constructing a systemic offer with shared interest, shared uncertainty and high economic, environmental and social impacts. Innovation management, ecosystem, and public-private partnership literatures have been investigated, as well as two European Commission funded research projects aimed at smart mobility infrastructure development. Results show these projects are both critical and disappointing for each player. We explain this contradiction of value perception showing that partners need such ecosystem projects to go forward and update their competences and roadmaps, but that the observed project management approach hampers the collectively built learning and the evolution of the strategic agenda of each partner. In conclusion, we define the concept of proto-ecosystem as an intermediary 'management object' for innovation management, and point out implications to manage such projects in order to unfold their whole potential. (10.1504/IJATM.2018.093413)
  DOI : 10.1504/IJATM.2018.093413
• DyBED: An Efficient Algorithm for Updating Betweenness Centrality in Directed Dynamic Graphs
  
  • Chehreghani Mostafa Haghir
  • Bifet Albert
  • Abdessalem Talel
  , 2018, pp.2114-2123.
• Quarante ans d’imagerie satellitaire radar
  
  • Nicolas Jean-Marie
  • Tupin Florence
  Revue Française de Photogrammétrie et de Télédétection, Société Française de Photogrammétrie et de Télédétection, 2018.
• A Smooth Primal-Dual Optimization Framework for Nonsmooth Composite Minimization
  
  • Tran-Dinh Quoc
  • Fercoq Olivier
  • Cevher Volkan
  SIAM Journal on Optimization, Society for Industrial and Applied Mathematics, 2018, 28 (1), pp.96-134. We propose a new first-order primal-dual optimization framework for a convex optimization template with broad applications. Our optimization algorithms feature optimal convergence guarantees under a variety of common structure assumptions on the problem template. Our analysis relies on a novel combination of three classic ideas applied to the primal-dual gap function: smoothing, acceleration, and homotopy. The algorithms due to the new approach achieve the best known convergence rate results, in particular when the template consists of only non-smooth functions. We also outline a restart strategy for the acceleration to significantly enhance the practical performance. We demonstrate relations with the augmented Lagrangian method and show how to exploit the strongly convex objectives with rigorous convergence rate guarantees. We provide numerical evidence with two examples and illustrate that the new methods can outperform the state-of-the-art, including Chambolle-Pock, and the alternating direction method-of-multipliers algorithms.
• Transcription of Spanish Historical Handwritten Documents with Deep Neural Networks
  
  • Granell Emilio
  • Chammas Edgard
  • Likforman-Sulem Laurence
  • Martínez-Hinarejos Carlos-D
  • Mokbel Chafic
  • Cirstea Bogdan
  Journal of Imaging, MDPI, 2018, 4 (1), pp.22.
• Mathematical models for very high resolution SAR data and their applications
  
  • Deledalle Charles-Alban
  • Denis L.
  • Ferraioli G.
  • Tupin Florence
  , 2018.
• Une approche par patchs, multi-atlas, itérative pour la segmentation du cortex cérébral en IRM néonatale
  
  • Tor-Díez Carlos
  • Passat Nicolas
  • Bloch Isabelle
  • Faisan Sylvain
  • Bednarek Nathalie
  • Rousseau François
  , 2018. L’analyse des structures cérébrales chez le nouveau-né constitue un enjeu de santé majeur, notamment en cas de prématurité, afin de disposer d’informations prédictives sur le développement de l’enfant. Le cortex est, en particulier, une structure d’intérêt, observable en IRM (imagerie par résonance magnétique). Les données IRM néonatales présentent toutefois des spécificités qui les rendent complexes à traiter. Dans ce contexte, les approches multi-atlas constituent une stratégie efficace, tirant parti de données traitées préalablement. La méthode proposée dans cet article repose sur une telle stratégie multi-atlas. Elle s’appuie notamment sur deux paradigmes : l’utilisation d’un modèle non local à base de patchs, et l’utilisation d’un schéma d’optimisation itératif. L’usage couplé de ces deux concepts permet notamment de considérer des patchs liés à l’image ainsi qu’à sa segmentation courante. Cette stratégie, comparée à de précédentes méthodes multi-atlas de la littérature, aboutit à des résultats de segmentation corticale robustes.
• Mass volume curves and anomaly ranking
  
  • Clémençon Stéphan
  • Thomas Albert
  Electronic Journal of Statistics, Shaker Heights, OH : Institute of Mathematical Statistics, 2018, 12 (2), pp.2806-2872. (10.1214/18-EJS1474)
  DOI : 10.1214/18-EJS1474
• Mean value coordinates for quad cages in 3D
  
  • Thiery Jean-Marc
  • Memari Pooran
  • Boubekeur Tamy
  ACM Transactions on Graphics, Association for Computing Machinery, 2018.
• Method, device and computer program for encapsulating media data into a media file
  
  • Denoual Franck
  • Mazé Frédéric
  • Le Feuvre J.
  • Ouedraogo Nael
  , 2018.
• Attack Tree Construction and Its Application to the Connected Vehicle
  
  • Danger Jean-Luc
  • Karray Khaled
  • Guilley Sylvain
  • Abdelaziz Elaabid M.
  , 2018, pp.175-190. (10.1007/978-3-319-98935-8_9)
  DOI : 10.1007/978-3-319-98935-8_9
• Segmentation of pelvic vessels in pediatric MRI using a patch based learning approach
  
  • Virzi Alessio
  • Gori Pietro
  • Muller Cécile
  • Mille Eva
  • Peyrot Quoc
  • Berteloot Laureline
  • Boddaert Nathalie
  • Sarnacki Sabine
  • Bloch Isabelle
  , 2018, pp.617.
• Musical Descriptions Based on Formal Concept Analysis and Mathematical Morphology
  
  • Agon Carlos
  • Andreatta Moreno
  • Atif Jamal
  • Bloch Isabelle
  • Mascarade Pierre
  , 2018, pp.105-119. In the context of mathematical and computational representations of musical structures, we propose algebraic models for formalizing and understanding the harmonic forms underlying musical compositions. These models make use of ideas and notions belonging to two algebraic approaches: Formal Concept Analysis (FCA) and Mathematical Morphology (MM). Concept lattices are built from interval structures whereas mathematical morphology operators are subsequently defined upon them. Special equivalence relations preserving the ordering structure of the lattice are introduced in order to define musically relevant quotient lattices modulo congruences. We show that the derived descrip-tors are well adapted for music analysis by taking as a case study Ligeti's String Quartet No. 2. (10.1007/978-3-319-91379-7_9)
  DOI : 10.1007/978-3-319-91379-7_9
• On the optimality and practicability of mutual information analysis in some scenarios
  
  • de Chèrisey Èloi
  • Guilley Sylvain
  • Heuser Annelie
  • Rioul Olivier
  Cryptography and Communications - Discrete Structures, Boolean Functions and Sequences, Springer, 2018, 10 (1), pp.101-121. The best possible side-channel attack maximizes the success rate and would correspond to a maximum likelihood (ML) distinguisher if the leakage probabilities were totally known or accurately estimated in a profiling phase. When profiling is unavailable, however, it is not clear whether Mutual Information Analysis (MIA), Correlation Power Analysis (CPA), or Linear Regression Analysis (LRA) would be the most successful in a given scenario. In this paper, we show that MIA coincides with the maximum likelihood expression when leakage probabilities are replaced by online estimated probabilities. Moreover, we show that the calculation of MIA is lighter that the computation of the maximum likelihood. We then exhibit two case-studies where MIA outperforms CPA. One case is when the leakage model is known but the noise is not Gaussian. The second case is when the leakage model is partially unknown and the noise is Gaussian. In the latter scenario MIA is more efficient than LRA of any order. (10.1007/s12095-017-0241-x)
  DOI : 10.1007/s12095-017-0241-x
• A contrario comparison of local descriptors for change detection in Very High spatial Resolution (VHR) satellite images of urban areas
  
  • Tupin Florence
  • Liu Gang
  • Gousseau Yann
  IEEE Transactions on Geoscience and Remote Sensing, Institute of Electrical and Electronics Engineers, 2018. Change detection is a key problem for many remote sensing applications. In this paper, we present a novel unsupervised method for change detection between two high resolution remote sensing images possibly acquired by two different sensors. This method is based on keypoints matching, evaluation and grouping, and does not require any image co-registration. It consists of two main steps. First, global and local mapping functions are estimated through keypoints extraction and matching. Secondly, based on these mappings, keypoint matchings are used to detect changes and then grouped to extract regions of changes. Both steps are defined through an {\it a contrario} framework, simplifying the parameter setting and providing a robust pipeline. The proposed approach is evaluated on synthetic and real data from different optic sensors with different resolutions, incidence angles and illumination conditions. (10.1109/TGRS.2018.2888985)
  DOI : 10.1109/TGRS.2018.2888985
• Prediction of weakly locally stationary processes by auto-regression
  
  • Roueff François
  • Sanchez-Perez Andres
  ALEA : Latin American Journal of Probability and Mathematical Statistics, Instituto Nacional de Matemática Pura e Aplicada (Rio de Janeiro, Brasil) [2006-....], 2018, 15, pp.1215–1239. In this contribution we introduce weakly locally stationary time series through the local approximation of the non-stationary covariance structure by a stationary one. This allows us to define autoregression coefficients in a non-stationary context, which, in the particular case of a locally stationary Time Varying Autoregressive (TVAR) process, coincide with the generating coefficients. We provide and study an estimator of the time varying autoregression coefficients in a general setting. The proposed estimator of these coefficients enjoys an optimal minimax convergence rate under limited smoothness conditions. In a second step, using a bias reduction technique, we derive a minimax-rate estimator for arbitrarily smooth time-evolving coefficients, which outperforms the previous one for large data sets. In turn, for TVAR processes, the predictor derived from the estimator exhibits an optimal minimax prediction rate. (10.30757/ALEA.v15-45)
  DOI : 10.30757/ALEA.v15-45
• High-Dimensional Mixture Models For Unsupervised Image Denoising (HDMI)
  
  • Houdard Antoine
  • Bouveyron Charles
  • Delon Julie
  SIAM Journal on Imaging Sciences, Society for Industrial and Applied Mathematics, 2018. This work addresses the problem of patch-based image denoising through the unsupervised learning of a probabilistic high-dimensional mixture models on the noisy patches. The model, named hereafter HDMI, proposes a full modeling of the process that is supposed to have generated the noisy patches. To overcome the potential estimation problems due to the high dimension of the patches, the HDMI model adopts a parsimonious modeling which assumes that the data live in group-specific subspaces of low dimension-alities. This parsimonious modeling allows in turn to get a numerically stable computation of the conditional expectation of the image which is applied for denoising. The use of such a model also permits to rely on model selection tools, such as BIC, to automatically determine the intrinsic dimensions of the subspaces and the variance of the noise. This yields a blind denoising algorithm that demonstrates state-of-the-art performance, both when the noise level is known and unknown.
• Pas de probas, pas de chocolat !
  
  • Zayana Karim
  Au fil des maths, APMEP, 2018. Expériences aléatoires, lois discrètes et continues, approximation des unes par les autres, intervalles de confiance, fluctuations d’échantillonnage, tests statistiques, paradoxes probabilistes
• Adaptive random forests for data stream regression
  
  • Gomes Heitor Murilo
  • Barddal Jean Paul
  • Ferreira Luis Eduardo Boiko
  • Bifet Albert
  , 2018.
• Online Learning with Reoccurring Drifts: The Perspective of Case-Based Reasoning
  
  • Al-Ghossein Marie
  • Murena Pierre-Alexandre
  • Cornuéjols Antoine
  • Abdessalem Talel
  , 2018.
• Gaussian Priors for Image denoising
  
  • Delon Julie
  • Houdard Antoine
  , 2018. This chapter is dedicated to the study of Gaussian priors for patch-based image denoising. In the last twelve years, patch priors have been widely used for image restoration. In a Bayesian framework, such priors on patches can be used for instance to estimate a clean patch from its noisy version, via classical estimators such as the conditional expectation or the maximum a posteriori. As we will recall, in the case of Gaussian white noise, simply assuming Gaussian (or Mixture of Gaussians) priors on patches leads to very simple closed-form expressions for some of these estimators. Nevertheless, the convenience of such models should not prevail over their relevance. For this reason, we also discuss how these models represent patches and what kind of information they encode. The end of the chapter focuses on the different ways in which these models can be learned on real data. This stage is particularly challenging because of the curse of dimensionality. Through these different questions, we compare and connect several denoising methods using this framework.