| lamsade.ml@gmail.com |  |
Thu 6th Jun 14 h 30 min 17 h 00 min | gr. de lecture Alex A. LDR D102 |  |
Fri 7th Jun 10 h 00 min 12 h 00 min | Arrivée Tom Darmon (stagiaire Causalité) | |
Thu 13th Jun 14 h 30 min 17 h 00 min | gr. de lecture Celine D102 | |
Thu 20th Jun 13 h 30 min 16 h 00 min | gr. de lecture Mateus Graph Neural networks D102 | |
Thu 27th Jun 14 h 30 min 17 h 00 min | gr. de lecture Alex V. (reseau inversible) D102 | |
Fri 28th Jun 14 h 00 min 15 h 00 min | séminaire de Aurelien Bellet room C131 Title: Privacy-Preserving Algorithms for Decentralized Collaborative Machine Learning Abstract: With the advent of connected devices with computation and storage capabilities, it becomes possible to run machine learning on-device to provide personalized services. However, the currently dominant approach is to centralize data from all users on an external server for batch processing, sometimes without explicit consent from users and with little oversight. This centralization poses important privacy issues in applications involving sensitive data such as speech, medical records or geolocation logs. In this talk, I will discuss an alternative setting where many agents with local datasets collaborate to learn models by engaging in a fully decentralized peer-to-peer network. We introduce and analyze asynchronous algorithms that allow agents to improve upon their locally trained model by exchanging information with other agents that have similar objectives. I will then describe how to make such algorithms differentially private to avoid leaking information about the local datasets, and analyze the resulting privacy-utility trade-off. These results are illustrated by a set of numerical experiments. | |
Wed 3rd Jul 11 h 00 min to Fri 5th Jul 20 h 00 min | CAP | |
Tue 3rd Sep 14 h 00 min 20 h 00 min | AI For Finance 2:00 pm - Damien GROMIER, CEO, Startup Inside - Founder, AI for Finance Mathieu ROUX, Executive Director, Palais Brongniart 2:05 pm - Cédric VILLANI, Mathematician, Field Medal Recipient Member of Parliament -expected* 2:15 pm - Philippe AYMERICH, Deputy CEO, Société Générale 2:25 pm - Charles-Edouard BOUÉE, CEO, Roland Berger 2:35 pm - Aldrick ZAPPELLINI, Data Director, Crédit Agricole SA 2:45 pm - AI & Cybersecurity David SADEK, VP Research, Technology and Innovation, Thales Christopher MUFFAT, CEO, Dathena Raphael DE CORMIS, VP Business Angel Cyber & AI, Thales 3:05 pm - Risk management use cases Adina GRIGORIU, CEO, Active Asset Allocation Sophie ELKRIEF, Chief Investment Offcer, MAIF Anne LAMOTTE, “My Future“ ecosystem leader, Allianz Emma SEZEN, Head of AI for Finance, Startup Inside 3:25 pm - Can AI accelerate local ecosystem’s growth? Jean-Paul MAZOYER, General Manager, Crédit Agricole Pyrénées Gascogne Carlo PURASSANTA, President of Microsoft France, Microsoft 3:45 pm - Customer experience with AI Gregory RENARD, Co-Founder and Chief AI Offcer, xBrain 3:55 pm - Insurance use case Jean-Laurent GRANIER, CEO, Generali France 4:05 pm - Magali NOE, CDO, CNP Assurances Gilles MOYSE, CEO, reciTAL 4:20 pm - Ronan LE MOAL, CEO, Arkea 4:35 pm - AI in retail Jean-Philippe DESBIOLLES, Vice President, IBM Watson Europe 4:35 pm - Jean-Philippe DESBIOLLES, Vice President, IBM Watson Europe Arnaud CAUDOUX, Executive Vice President, BPI France Claire CALMEJANE, Group CIO, Société Générale Marie-Caroline BAERD, Executive VP AI Offer Leader, Capgemini Invent Cyrille FOILLARD, CIO, RCI Bank and Services Damien PERILLAT, Managing Director Western Europe, PayPal 5:10pm - From POC to industrialization of AI use cases David GIBLAS, CDO, Malakoff Mederic Humanis Pascal COGGIA, CEO UK, ARTEFACT Lorenzo CROATI, VP and Co-Founder, Startup Inside 5:25 pm- Ethics by design in AI Bibi NDIAYE, Innovation Director and Data Intelligence, BPCE Jean-Marc BONNEFOUS, Co-Founder and Board member, Bonseyes Frère Eric SALOBIR, President, Optic Technology Alex PANICAN, Head of Partnerships and Ecosystem, The LHOFT 5:45pm - Cloud Migration Eric HADDAD, Managing Director, Google Cloud Arnaud MULLER, CEO & Founder, Saagie 6:05pm - Laurent MIGNON, CEO and Chairman of Group BPCE Management Board, BPCE 6:20 pm - Deploying trusted and industrialized AI Philippe LIMANTOUR, CIO, EY 6:30pm - Renaud DUMORA, CEO, BNP Paribas Cardif 6:45pm - AI for green fnance Emmanuel BACRY, Chief Scientifc Offcer, Health Data Hub Grégory LABROUSSE, CEO & Founder, NamR Bettina LAVILLE, Conseiller d'Etat 7:00 pm- Reinforcement learning for algorithmic trading Charles-Albert LEHALLE, Head of Data Analytics and visiting researcher, CFM & Imperial College London 7:10pm - Using deep learning to deliver value to banks and their customers Imtiaz ADAM, CEO Founder & Director of DATA, Deep Learn Strategies 7:20pm - Fraud Detection Yannick MARTEL, Co founder & Head of strategy, Bleckwen Caroline LAMAUD, Chairman & Co-Founder, Anaxago Remy SANONER, Senior Advisor, Startup Inside 7:30pm - AI & quantum computing Paula FORTEZA, In charge of the report on the Quantum Technologies, Member of the French Parliament, 7:40pm - Pierre-Alain RAPHAN, Member of the French Parliament 7:50pm - Agnès PANNIER-RUNACHER, Secretary of State to the Minister of Economy and Finance -expected Palais Brongniart, 16 Place de la Bourse, 75002 Paris, France  | |
Thu 19th Sep 15 h 00 min 17 h 00 min | groupe de lecture Eric on neural hawkes process papers.nips.cc/paper/7.....s.pdf salle B 113 | |
Thu 26th Sep 10 h 30 min 11 h 30 min | Séminaire pole 3 - Fabian Castano Nous allons avoir le plaisir d'écouter Fabian Castaño, Assistant Professor à Ponticia Universidad Javeriana, Colombie (invité du 23 au 27 septembre au LAMSADE par André Rossi), nous présenter ses travaux lors du prochain séminaire du pole 3. Title : Using open access data to model a technician routing and scheduling problem in a congested urban setting Date : September 26 - 10h30 Room : B208 Abstract: This research aims at studying and improving the attended home services delivery in the city of Bogota (Colombia) by considering varying traffic patterns along the day. The goal is to improve routing decisions by adopting appropriate travel time functions between pairs of locations that are built upon the basis of freely available information gathered from collaborative consumption platforms. In first place, data collected from Uber Movement ® platform is used to identify the 1-hour consolidated traffic patterns based on coordinates and time of the day for all pairs of neighborhoods within the city. However, not all the information is available, as there might exist some journeys scarcely required by Uber users. To overcome this problem, a K-nearest neighbor regression is used to predict missing travel times when required and considering as an input the coordinates and the time of the day. Then, a piecewise linear function representing the travel time between pairs of locations is constructed by assuming that a breakpoint, or change on travel time behavior, takes place at the middle point of each one-hour time frame. A set of piecewise linear functions is then obtained after solving a system of linear equations. Following, two different approaches are used to generate solutions for a technician routing and scheduling problem with hard time windows. The first consists in an approximation Integer Programming model that represents the problems using an acyclic directed graph. The second consist in a Memetic Algorithm that minimizes the number of vehicles and vehicles field time. Preliminary results show that both approaches present similar results and can be used in practical applications. Pour en savoir plus : www.javerianacali.edu......stano B208 | |
Thu 26th Sep 15 h 00 min 16 h 30 min | groupe de lecture miles Eric on Harmonic Exponential Families on Manifolds tacocohen.files.wordpr.....f.pdf ICML 2015 Dauphine, B 113 | |
Thu 10th Oct 8 h 30 min 10 h 30 min | Séminaires ED Manel Ayadi + Alexandre ARAUJO | |
Thu 10th Oct 11 h 00 min 12 h 00 min | Séminaire pole 3 - Ryoma Sato Annonce à envoyer Ryoma Sato - Kyoto University / RIKEN AIP Title: Machine Learning for Combinatorial Graph Problems Abstract: Combinatorial graph problems are everywhere in computer science. For example, the scheduling problem can be formulated as the graph coloring problem, and the community detection problem can be formulated as the clique problem. Traditionally, algorithms for solving these problems have been invented by domain experts. Recently, machine learning methods have succeeded in many applications on graph data, such as community detection in social networks and automatic drug discovery. In this talk, I will describe the utilities and limitations of machine learning methods for solving combinatorial problems. First, I will introduce how to discover the worst-case instance of combinatorial problems using machine learning methods. This method helps us design efficient algorithms and understand the behavior of algorithms. Next, I will introduce the theoretical limitations of graph neural networks for solving combinatorial problems. Graph neural networks are the state-of-the-art machine learning method for graph data, and we expected that graph neural networks could learn efficient algorithms for solving combinatorial graph problems. However, we prove the graph neural networks cannot learn better algorithms than simple greedy algorithms in terms of approximation ratios. This talk is based on these two papers: * Ryoma Sato, Makoto Yamada, Hisashi Kashima, Learning to Sample Hard Instances of Graph Problems, ACML 2019. * Ryoma Sato, Makoto Yamada, Hisashi Kashima, Approximation Ratios of Graph Neural Networks for Combinatorial Problems, NeurIPS 2019. ``` A309 | |
Thu 17th Oct 9 h 00 min 18 h 00 min | gdr ISIS - théorie du deep learning Réunion du GdR ISIS Titre : Théorie du deep learning Dates : 2019-10-17 Lieu : Cnam Paris, amphithéâtre Paul Painlevé - 292, rue Saint Martin 75003 Paris. Annonce : Les réseaux de neurones profonds ont marqué l'entrée dans une nouvelle ère de l'intelligence artificielle, ponctuée par des succès opérationnels dans des domaines variés de la science des données comme la classification d'images, la reconnaissance vocale, ou le traitement de la langue naturelle. En dépit de ces succès importants, les garanties théoriques associées à ces modèles décisionnels restent aujourd'hui toujours fragiles. L'objectif de cette journée est de faire un état des lieux sur la compréhension du fonctionnement des réseaux de neurones profonds, à travers un appel à contributions centré autour les thèmes (non exhaustifs) suivants : Expressivité des modèles Robustesse décisionnelle (incertitude, stabilité, attaques adversaires) Optimisation et problèmes non convexes Théorie de la généralisation Lien entre modèles physiques et architectures de réseaux de neurones Les outils utilisés pour aborder ces thématiques pourront venir de l'apprentissage statistique, mais des méthodes venant de disciplines connexes (décomposition tensorielles, analyse harmonique, méthodes géométriques / algébriques, physique statistique) sont fortement encouragées. Orateurs inivtés : Rémi Gribonval, LIP ENS Lyon, Edouard Oyallon, LIP6 Paris, CNRS Appel à contributions : Les personnes souhaitant présenter leurs travaux à cette journée sont invitées à envoyer, par e-mail, leur proposition (titre et résumé de 1 page maximum) aux organisateurs avant le 26 septembre 2019. Organisateurs : Caroline Chaux-Moulin (valentin.emiya@lis-lab.fr), Université Aix-Marseille, I2M Valentin Emiya (caroline.chaux@univ-amu.fr), Université Aix-Marseille, LIS François Malgouyres (Francois.Malgouyres@math.univ-toulouse.fr), Institut de Mathématiques de Toulouse (IMT, CNRS UMR 5219) Nicolas Thome (nicolas.thome@cnam.fr), Cnam Paris Konstantin Usevich (konstantin.usevich@univ-lorraine.fr), Université de Lorraine, CRAN, Nancy Lien : gdr-isis.fr/index.php?p.....n=405 Cnam Paris, amphithéâtre Paul Painlevé - 292, rue Saint Martin 75003 Paris | |
Thu 17th Oct 13 h 00 min 15 h 00 min | Open Science: LAMSADE | |
Thu 17th Oct 15 h 00 min 16 h 30 min | SMILE seminar - topic: double descent phenomenon | |
Thu 7th Nov 10 h 00 min 12 h 00 min | Séminaires ED Eric Benhamou + Raphael Pinot | |
Thu 21st Nov 10 h 45 min 11 h 45 min | Séminaire pole 3 - Arthur Mensch Title : Functional smoothing for sparse and cost-informed prediction of output distributions. Date : November 21 - 10h45 Room : B413 Abstract: To facilitate training with gradients, supervised learning methods often transform selecting a single element within a set of outputs to predicting a probability distribution over this set (using e.g. the softmax operator). In this talk, we will understand this transformation as a functional smoothing of the output selection mechanism. Engineering this Nesterov smoothing yields new modelling perspective. First, we will observe that selecting an output within a combinatorial set (e.g. a sequence of tags) is often solved using dynamic programming algorithms. Smoothing turn DP algorithms into differentiable operators, that may predict potentially sparse probabilities over the output set. Secondly, we will design a smoothing that takes into account a cost function defined on the output set. This approach transforms the softmax operator into a cost-informed geometric softmax, that has the further capabilities of predicting distributions over a continuous set. Pour en savoir plus : www.amensch.fr B413 | |
Thu 28th Nov 10 h 45 min 11 h 45 min | Séminaire pole 3 - Nicolas Usunier Title : Fair machine learning: optimal predictions under parity constraints Date : November 28 - 10h45 Room : A303 Abstract:There is increasing attention given to the question of whether automated decision systems based on machine learning are unfair, or biased, with respect to protected characteristics such as gender or ethnicity. For classification problems, a common fairness criterion is demographic parity (DP), which states that if X% of the population belongs to one group, then X% of the individuals receiving a favorable outcome should belong to that group. In this talk, I will briefly motivate the main fairness criteria used in machine learning, and present our study on optimal predictors under parity constraints (DP, and equal opportunity (EO)) in a general setting including classification, regression and probability estimation. We give closed-form formula of Bayes-optimal predictors, which improves the understanding of the effects of parity constraints. Our results stress the predominant role of the rank of individuals within their protected group, already known in classification, in the problems of fair regression and probability estimation. We also show that under the DP (or EO) constraints, the optimal classifiers are obtained by thresholding the optimal estimator of posterior probability, as in the usual unconstrained case, whereas all previous results use group-dependent thresholds. As a by-product, we give analytical formula of optimal fair representations, learned with the intention to obfuscate the protected attribute while preserving the most relevant information. Our analysis stresses the difficulty of the unsupervised learning of such fair representations. Joint work with Nicolas Carion, Moustapha Cisse and Yves Grandvalet Short bio: Nicolas Usunier is Research Scientist at Facebook AI Research since 2015. He was previously Associate professor at Université de Technologie de Compiègne (UTC), and Assistant professor at Université Paris VI. His research interests include machine learning theory, reinforcement learning and fairness in machine learning. A303 | |
Thu 16th Jan 10 h 00 min 12 h 00 min | Seminaire ED Amin Farvardin + | |
Thu 23rd Jan 11 h 00 min 12 h 00 min | groupe de lecture MILES Oratrice: virginie Do P516 Inherent Trade-Offs in the Fair Determination of Risk Scores arxiv.org/abs/1609.05807 P516 | |
Mon 27th Jan 15 h 30 min 16 h 30 min | GdT STAT-NUM Notre collègue Zhenjie Ren nous parlera de "Game on random environment and Mean-field Langevin system with application to GAN" | |
Thu 13th Feb 10 h 00 min 12 h 00 min | Séminaires ED : Raja Trabelsi + Felippe Garrido | |
Thu 12th Mar 10 h 00 min 12 h 00 min | Séminaires ED Hajer Ben Fekih + Ons Nefla | |
Thu 14th May 10 h 00 min 12 h 00 min | Séminaires ED Geovani Rizk+ Beji Celine | |
Thu 18th Jun 10 h 00 min 12 h 00 min | Séminaires ED Axel Faure Beaulieu + | |