Upload
trinhkhuong
View
220
Download
2
Embed Size (px)
Citation preview
lamsade
Machine Learning
Machine Learning
Florian Yger
HCERES Visit, 10/24/17
Florian Yger Machine Learning
lamsade
Machine Learning
Themes of the project
Overall aimHow to learn optimal decisions from large amounts of data in achanging context ?
Scientific challenges - at the cross-road between pole 1 and pole 3
Representation learning (and invariance design)Learning for decision making under uncertainty
Florian Yger Machine Learning
lamsade
Machine Learning
Themes of the project
Scientific challenges - at the cross-road between pole 1 and pole 3
Representation learning (and invariance design)Learning for decision making under uncertainty
Florian Yger Machine Learning
lamsade
Machine Learning
Themes of the project
Scientific challenges - at the cross-road between pole 1 and pole 3
Representation learning (and invariance design)Learning for decision making under uncertainty
Florian Yger Machine Learning
lamsade
Machine Learning
Themes of the project
Scientific challenges - at the cross-road between pole 1 and pole 3
Representation learning (and invariance design)Learning for decision making under uncertainty
ApplicationsGamesBiomedical signal processingRecommender systems,...
Florian Yger Machine Learning
lamsade
Machine Learning
Human resources
Permanent staff
J. Atif (PU, 100%) - 2014T. Cazenave (PU, 50%) - 2009Y. Chevaleyre (PU, 100%) - 2017R. Laraki (DR, 20%) - 2013B. Negrevergne (MCF, 100%) - 2016F. Yger (MCF, 100%) - 2015
Florian Yger Machine Learning
lamsade
Machine Learning
Human resources
International collaborations with: Google Brain NY, Riken AIP, KU Leuven, etc
Florian Yger Machine Learning
lamsade
Machine Learning
Human resources
Non-permanent staffM. Cornu (2013 - , MESR)F. Labernia (2014-2017, MESR)S. Hadikhanloo (2014 - , AD/CEREMADE)A. Morvan (2015 - , CEA)C. Béji (2016 - , MESR)B. Doux (2017 - , MESR)R. Pinot (2017 - , CEA)Y. Laurin (2017 - , ENS)S. Nicolet (2017 - , PRAG)P. Meriguet (2017 - , Cifre Smith detection/CEREMADE)A. Araujo (2017 - , Cifre Wavestone)N. Carion (2017 - , Cifre Facebook)T. Sohm-Quéron (2017 - , Cifre Brabham gardens)K. Osanlou (2017 - , Cifre Safran)G. Sileno (2016 - , postdoc fellow)M. Bon (2017 - , industrial associate)
Florian Yger Machine Learning
lamsade
Machine Learning
Human resources
Selection of thesis topicsF. Labernia, Modelisation, learning and preference prediction,(J. Atif)C. Beji, Uplift modelling and causal inference, (J. Atif, F. Yger)R. Pinot, Online sketching of structured data for machinelearning under differential privacy, (J. Atif, F. Yger)A. Araujo, Dynamical optimization of machine learningapplications, (J. Atif, B. Negrevergne)B. Doux, Detecting and understanding salient events in videogame data, (T. Cazenave, B. Negrevergne)S. Hadikhanloo, Learning in Mean Field Games, (R. Laraki)
Florian Yger Machine Learning
lamsade
Machine Learning
Other resources
Academic projects:national level
ANR Logima (2014-2017), 72 Keuros, Partner, Consortium:TPT, ECP, Univ. Dauphine.ANR ESIGMA (2017-2021), 141Keuros, Partner, Consortium:École polytechnique, Dauphine, Univ. Montpellier.PEPS-JCJC ADDICTED (2017), 10Keuros
international levelANR STAP (2017-2021, 137K), Partner, Consortium: TPT,Univ. Dauphine, IME Brazil
Industrial projects:Adway (2015-2016) (2017- ) 25Keuros/yearCIFRE (Brabham gardens, Facebook, Safran, Smith detection,Wavestone)CEA (thesis fundings)
Florian Yger Machine Learning
lamsade
Machine Learning
Some achievements
Representation learning
Y. Isaac, Q. Barthélemy, C. Gouy-Pailler, M. Sebag, J. Atif.Multi-dimensional signal approximation with sparse structured priors usingsplit bregman iterations. Signal Processing, 2017.
M. Bojarski, A. Choromanska, K. Choromanski, F. Fagan, C. Gouy-Pailler,A. Morvan, N. Sakr, T. Sarlos, J. Atif. Structured adaptive and randomspinners for fast machine learning computations. In AISTATS 2017.
A. Lecoutre, B. Negrevergne, F. Yger. Recognizing Art StyleAutomatically in painting with deep learning, ACML, 2017
Selection among 20 international journal articles and38 international conference articles.
Florian Yger Machine Learning
lamsade
Machine Learning
Some achievements
Learning and decision under uncertainty
T. Cazenave. Generalized rapid action value estimation. IJCAI, 2015.
J. Flesch, R. Laraki, V. Perchet. Online learning and blackwellapproachability in quitting games. COLT 2016.
B. Negrevergne, T. Cazenave. Distributed nested rollout policy for samegame. In Computer Games Workshop at IJCAI. 2017.
Selection among 20 international journal articles and38 international conference articles.
Florian Yger Machine Learning
lamsade
Machine Learning
Some achievements
Representation to cope with non-stationarity /changing contexts
F. Labernia, B. Zanuttini, B. Mayag, F. Yger, J. Atif. Online learning ofacyclic conditional preference networks from noisy data, IEEE ICDM2017.
T. Cazenave. Residual networks for computer go. IEEE Transactions onComputational Intelligence and AI in Games, 2017.
I. Horev, F. Yger, and M. Sugiyama. Geometry-aware stationary subspaceanalysis. ACML, 2016.
Selection among 20 international journal articles and38 international conference articles.
Florian Yger Machine Learning
lamsade
Machine Learning
Perspectives and scientific challenges
Summaryrecent project in rapid expansion,great quality and quantity of international collaborations andpublications.
Participation to society’s next challengesincorporation of privacy constraintsexplainability of the decisions
Directions for 2017-2022reinforcement of the transversality of the projectwithin LAMSADE, Dauphine, PSLcapitalization on the existing collaborations (national andinternational)
Florian Yger Machine Learning