Publications

2017

• Sampling from a log-concave distribution with compact support with proximal Langevin Monte Carlo
  
  • Brosse Nicolas
  • Durmus Alain
  • Moulines Éric
  • Pereyra Marcelo
  Proceedings of Machine Learning Research, PMLR, 2017, 65, pp.319-342. This paper presents a detailed theoretical analysis of the Langevin Monte Carlo sampling algorithm recently introduced in [DMP16] when applied to log-concave probability distributions that are restricted to a convex body K. This method relies on a regularisation procedure involving the Moreau-Yosida envelope of the indicator function associated with K. Explicit convergence bounds in total variation norm and in Wasserstein distance of order 1 are established. In particular, we show that the complexity of this algorithm given a first order oracle is polynomial in the dimension of the state space. Finally, some numerical experiments are presented to compare our method with competing MCMC approaches from the literature.
• Fair throughput allocation in Information-Centric Networks
  
  • Bonald Thomas
  • Mekinda Léonce
  • Muscariello Luca
  Computer Networks, Elsevier, 2017, 125, pp.122 - 131. Cache networks are the cornerstones of today's Internet, helping it to scale by an extensive use of Content Delivery Networks (CDN). Benefiting from CDN's successful insights, ubiquitous caching through Information-Centric Networks (ICN) is increasingly regarded as a premier future Internet architecture contestant. However, the use of in-network caches seems to cause an issue in the fairness of resource sharing among contents. Indeed, in legacy communication networks, link buffers were the principal resources to be shared. Under max-min flow-wise fair bandwidth sharing [14], content throughput was not tied to content popularity. Including caches in this ecosystem raises new issues since common cache management policies such as probabilistic Least Recently Used (p-LRU) or even more, Least Frequently Used (LFU), may seem detrimental to low popularity objects, even though they significantly decrease the overall link load [3]. In this paper, we demonstrate that globally achieving LFU is a first stage of content-wise fairness. Indeed, any investigated content-wise α-fair throughput allocation permanently stores the most popular contents in network caches by ensuring them a cache hit ratio of 1. As ICN caching traditionally pursues LFU objectives, content-wise fairness specifics remain only a matter of fair bandwidth sharing, keeping the cache management intact. (10.1016/j.comnet.2017.05.019)
  DOI : 10.1016/j.comnet.2017.05.019
• A user-perception based approach to create smiling embodied conversational agents
  
  • Ochs Magalie
  • Mckeown Gary
  • Pelachaud Catherine
  ACM Transactions on Interactive Intelligent Systems, Association for Computing Machinery (ACM), 2017, 1. no abstract
• Predicting Completeness in Knowledge Bases
  
  • Galárraga Luis
  • Razniewski Simon
  • Amarilli Antoine
  • Suchanek Fabian M.
  , 2017. no abstract
• Quadratic Extension Field Codes for Free Space Optical Intensity Communications
  
  • Mroueh L.
  • Belfiore Jean-Claude
  IEEE Transactions on Communications, Institute of Electrical and Electronics Engineers, 2017, 65 (2), pp.751 - 763.
• Failure Analysis in Magnetic Tunnel Junction Nanopillar with Interfacial Perpendicular Magnetic Anisotropy
  
  • Zhao Weisheng
  • Wang You
  • Naviner Lirida
  Materials Science Journal, 2017, 9 (41), pp.1-17.
• Guest Editorial AWPL Special Cluster on “Impact of User-Related Randomness on Antennas and Channels”
  
  • Sibille Alain
  • Kildal Per-Simon
  IEEE Antennas and Wireless Propagation Letters, Institute of Electrical and Electronics Engineers, 2017, 16. The guest editorial explains the motivation for the AWPL special cluster and briefly introduce each of the nine selected papers. (10.1109/LAWP.2017.2696621)
  DOI : 10.1109/LAWP.2017.2696621
• Fast algebraic immunity of Boolean functions
  
  • Mesnager Sihem
  • Cohen Gérard
  Advances in Mathematics of Communications, AIMS, 2017, 11 (2), pp.373-377. (10.3934/amc.2017031)
  DOI : 10.3934/amc.2017031
• LISP EID Block Management Guidelines
  
  • Iannone Luigi
  • Jorgensen Roger
  • Conrad David
  • Huston Geoff
  , 2017.
• Mesure de dissimilarité pour les patchs utilisant la corrélation
  
  • Riot Paul
  • Almansa Andrés
  • Gousseau Yann
  • Tupin Florence
  , 2017.
• White matter hyperintensities segmentation in a few seconds using fully convolutional network and transfer learning
  
  • Xu Yongchao
  • Géraud Thierry
  • Puybareau Elodie
  • Bloch Isabelle
  • Chazalon Joseph
  , 2017, LNCS. In this paper, we propose a fast automatic method that seg- ments white matter hyperintensities (WMH) in 3D brain MR images, using a fully convolutional network (FCN) and transfer learning. This FCN is VGG, pre-trained on ImageNet for natural image classification, and fine tuned with the training dataset of the MICCAI WMH Chal- lenge. We consider three images for each slice of volume to segment: the i-th T1 slice, the i-th FLAIR slice, and the residue of a morphological operator that emphasizes small bright structures. These three 2D images are assembled to form a 2D color image, that inputs the FCN to obtain the 2D segmentation of the i-th slice. We process all slices, and stack the results to form the 3D output segmentation. With such a technique, the segmentation of WMH on a 3D brain volume takes about 10 seconds. Our technique was ranked 6-th over 20 participants at the MICCAI WMH Challenge.
• Efficient Smoothed Concomitant Lasso Estimation for High Dimensional Regression
  
  • Ndiaye Eugene
  • Fercoq Olivier
  • Gramfort Alexandre
  • Leclère Vincent
  • Salmon Joseph
  Journal of Physics: Conference Series, IOP Science, 2017, J. Phys.: Conf. Ser. 904 012006. In high dimensional settings, sparse structures are crucial for efficiency, both in term of memory, computation and performance. It is customary to consider 1 penalty to enforce spar-sity in such scenarios. Sparsity enforcing methods, the Lasso being a canonical example, are popular candidates to address high dimension. For efficiency, they rely on tuning a parameter trading data fitting versus sparsity. For the Lasso theory to hold this tuning parameter should be proportional to the noise level, yet the latter is often unknown in practice. A possible remedy is to jointly optimize over the regression parameter as well as over the noise level. This has been considered under several names in the literature: Scaled-Lasso, Square-root Lasso, Concomitant Lasso estimation for instance, and could be of interest for confidence sets or uncertainty quantification. In this work, after illustrating numerical difficulties for the Smoothed Concomitant Lasso formulation, we propose a modification we coined Smoothed Concomitant Lasso, aimed at increasing numerical stability. We propose an efficient and accurate solver leading to a computational cost no more expansive than the one for the Lasso. We leverage on standard ingredients behind the success of fast Lasso solvers: a coordinate descent algorithm, combined with safe screening rules to achieve speed efficiency, by eliminating early irrelevant features. (10.1088/1742-6596/904/1/012006)
  DOI : 10.1088/1742-6596/904/1/012006
• Cognitive Management of Self -Organized Radio Networks Based on Multi Armed Bandit
  
  • Daher Tony
  • Jemaa Sana Ben
  • Decreusefond Laurent
  , 2017. Many tasks in current mobile networks are automated through Self-Organizing Networks (SON) functions. The actual implementation consists in a network with several SON functions deployed and operating independently. A Policy Based SON Manager (PBSM) has been introduced to configure these functions in a manner that makes the overall network fulfill the operator objectives. Given the large number of possible configurations (for each SON function instance in the network), we propose to empower the PBSM with learning capability. This Cognitive PBSM (C-PBSM) learns the most appropriate mapping between SON configurations and operator objectives based on past experience and network feedback. The proposed learning algorithm is a stochastic multi-armed bandit, namely the UCB1. We evaluate the performances of the proposed C-PBSM on an LTE-A simulator. We show that it is able to learn the optimal SON configuration and quickly adapts to objective changes.
• Raman-tailored photonic crystal fiber for telecom band photon-pair generation
  
  • Cordier Martin
  • Orieux Adeline
  • Gabet Renaud
  • Harlé Thibault
  • Dubreuil Nicolas
  • Diamanti Eleni
  • Delaye Philippe
  • Zaquine Isabelle
  Optics Letters, Optical Society of America - OSA Publishing, 2017, 42 (13), pp.2583-2586. We report on the experimental characterization of a novel nonlinear liquid-filled hollow-core photonic crystal fiber for the generation of photon pairs at a telecommuni- cation wavelength through spontaneous four-wave mixing (SFWM). We show that the optimization procedure in view of this application links the choice of the nonlinear liquid to the design parameters of the fiber, and we give an example of such an optimization at telecom wavelengths. Combining the modeling of the fiber and classical charac- terization techniques at these wavelengths, we identify for the chosen fiber and liquid combination SFWM phase- matching frequency ranges with no Raman scattering noise contamination. This is a first step toward obtaining a tele- com band fibered photon-pair source with a high signal-to- noise ratio. (10.1364/OL.42.002583)
  DOI : 10.1364/OL.42.002583