Seminars

The page below lists the coming and past seminars, and provides a link to the presentations that you may have missed. Click on a presentation title for the abstract.

Alert emails are sent to the TAU team and to the announcement mailing-list tau-seminars to which anyone can subscribe by clicking here. NB: if you do not receive a confirmation by email when you try to subscribe, please contact me directly.

The seminars take place on Tuesday afternoons at 14h30 in room 2014 (building 660), and are broadcasted online at https://bbb2.lri.fr/b/gui-hfj-kdg, unless specified otherwise.
Presentations are recorded and available here for the ones before 2021, while 2021 recordings are directly indicated inline.

March

• Tuesday, 14th of March, 14h30, Pierre Wolinsky (Statify team, Inria Grenoble-Alpes) : Gaussian Pre-Activations in Neural Networks: Myth or Reality? (Slides:Pres_ImposeGaussianPreactivations.pdf )

February

• Tuesday, 28th of February, 14h30, Filippo Masi (University of Sydney) : Thermodynamics-based Artificial Neural Networks (Slides:TAU_seminar_Masi.pdf )
• Tuesday, 21th of February, 14h30, Yulia Gusak (Inria Bordeaux) :Efficient Deep Learning
• Tuesday, 14th of February, 14h30, Beatriz Seoane Bartolomé(Departamento de Física Teórica,Un. Complutense de Madrid) :Explaining the effects of non-convergent sampling in the training of Energy-Based Models (Slides:LISN_BSeoane.pdf )

January

• Tuesday, 24th of January, 14h30, Bruno Loureiro (Center for Data Science, ENS Paris) : Phase diagram of Stochastic Gradient Descent in high-dimensional two-layer neural networks(Slides:slides_Loureiro_Bruno.pdf )
• Tuesday, 17th of January, 14h30, Vincenzo Schimmenti (Phd, TAU) : Assessing the predictive power of GPS data for aftershock pattern prediction

December

• Tuesday, 6th of December, 14h15, Alexander Reisach,Nicolas Atienza,Armand Lacombe, Shiyang Yan (TAU & AO team) : Causal learning project in the TAU team - Part 2 (Slides: TAU_seminars___06_12_22___Causal_Inference_projects.pdf)

November

• Tuesday, 29th of November, 14h15, Marylou Gabrié : Opportunities and Challenges in Enhancing Sampling with Learning (Slides: LRI_vsent.pdf)
• Tuesday, 22th of November, 14h15, Cyriaque Rousselot, Shuyu Dong and Audrey Poinsot (TAU & AO team) : Causal learning project in the TAU team - Part 1 (Slides: TAU_seminars___22_11_22___Causal_Inference_projects.pdf)
• Tuesday, 15th of November, 14h15, Manon Verbockhaven (TAU & AO team) : SPOTTING EXPRESSIVITY BOTTLENECKS AND FIXING THEM OPTIMALLY

October

• Tuesday, 25th of October, 14h15, Cyril Furtlehner (TAU) : Free Dynamics of Feature Learning Processes(Slides:Tau.pdf )
• Tuesday, 18th of October, 14h15, Pr. Vander Alves (University of Brasilia) : On the Interplay between Software Product Lines and Machine Learning Models
• Tuesday, 11th of October, 14h15, Pr. Li Weigang (University of Brasilia) : New Achievements of Artificial Intelligence in Multimodal Information Processing

July

• Monday, 5th of July, 14h30, Aymeric Blot (UCL) : Optimised source code as a Service

June

• Tuesday, 21th of June, 14h30, François Landes (TAU team) : Vocabulary and main stakes of fluid mechanics for dummies/for machine-learners
• Tuesday, 14th of June, 14h30, Yérali Gandica (LPTM, Cergy-Pontoise Univ.) : A Complex Systems approach to the emergence of socio-economic phenomena
• Tuesday, 7th of June, 14h30, Herilalaina Rakotoarison (TAU) : Learning meta-features for AutoML (link to paper https://openreview.net/forum?id=DTkEfj0Ygb8)

May

• Tuesday, 17th of May, 14h30, Martin Weigt (Sorbonne Université, Computational and Quantitative Biology) : Generative modeling of protein and RNA sequence ensembles
• Tuesday, 10th of May, 14h30, Yufei Han (Inria Renne-Bretagne-Atlantique) : Towards Understanding the Robustness Against Evasion Attack on Categorical Data

April

• Tuesday, 26th of April, 14h30, Jeremie Cabessa (LEMMA, Un. Paris 2) : Finite state machines and bio-inspired neural networks
• Tuesday, 19th of April, 14h30, Marin Ferecatu (Equipe Vertigo, Laboratoire CEDRIC, CNAM) : Méthodes d'apprentissage statistique pour l'analyse et l'exploration interactive des contenus visuels / Machine learning methods for analysis and interactive exploration of visual data
• Tuesday, 12th of April, 14h30, Sylvain Chevallier(LISV - IUT Vélizy - UVSQ - Univ. Paris-Saclay) : Learning invariant representations, application to anomaly detection and transfer learning for time series
• Monday, 4th of April, 11h00, Michele Buzzicotti(Dept. of Physics and INFN, University of Rome) : AI meets turbulence: Lagrangian and Eulerian data-driven tools for optimal navigation and data-assimilation

February

• Tuesday, 11th of January, 14h30, Olivier Goudet (Angers University): Population-based gradient descent weight learning for graph coloring problems
• Tuesday, 1st of January, 14h30, online: Olivier Teytaud (Facebook FAIR): Evolutionary Compilation and Baptiste Rozière (FAIR/Paris-Dauphine): Machine Learning for Source Code Translation

November

• Tuesday, 7th of November, 14:30 in room 2014 (building 660) and also online: Titouan Vayer (ENS Lyon) : Learning on incomparable spaces using Optimal Transport → recording
  Abstract

  Abstract: How to learn from multiple graphs or images of different resolutions? In this talk, I will present some strategies based on Optimal Transport theory (OT) which allows to define geometric notions of distance between probability distributions and to find correspondences, relations, between sets of points. I will present the Gromov-Wassertsein (GW) framework and show how it can be useful for dealing with structured data such as graphs. More specifically, I will address the problem of online graphs dictionary learning and show how to learn interesting graph representations with GW. On the image side, I will present the problem of CO Optimal Transport (COOT) whose goal is to align both samples and features of two different datasets. I will present applications of COOT in heterogeneous domain adaptation and co-clustering/data summarization.
  

October

• Tuesday, 16th of October, 11h30, in room 2014 and also online: Bruno Loureiro (EPFL) Exactly solvable models for high-dimensional inference and machine learning problems → recording
  Abstract

  In this talk I will discuss two lines of work that leverage the methods from Statistical Physics to study high-dimensional problems of interest in statistics and machine learning. The first line concerns the analysis of the algorithmic limitations of estimation in random inverse / inference problems. In particular, I will show how a class of generative models based on deep neural networks can present algorithmic advantages over the more classic sparse priors as regularisers for these inverse problems. The second line concerns the modelling of structured features for typical-case analysis of learning problems. I will introduce a Gaussian covariate model encompassing different tasks of interest in Machine Learning, and will discuss how its asymptotic performance can be characterised in high-dimensions. Finally, I will show that due to an interesting concentration phenomena, this model can be used to characterise the learning curves of realistic data, and some times even of real data.
  
  The talk will be mostly based on the following references: https://arxiv.org/abs/1905.12385, https://arxiv.org/abs/2102.08127
  
• Tuesday, 10th of October, 14h30, in room 445 "Patio", building 650 and also online: Tony Bonnaire (Institut d'Astrophysique Spatiale, Université Paris-Saclay): Learning patterns from point-cloud datasets and applications to cosmology → recording
  Abstract

  Point-cloud datasets are ubiquitous in many science and non-science fields. These data are usually coming along with unique patterns that some algorithms are meant to extract and that are linked with the underlying phenomenon that generated the data.In this presentation, we will focus on two kinds of spatially structured datasets. First, clustered-type patterns in which the datapoints are separated in the input space into multiple groups. We will show that the unsupervised clustering procedure performed with a Gaussian Mixture Model can be formulated in terms of a statistical physics optimisation problem. This formulation enables the unsupervised extraction of many key information about the dataset itself, like the number of clusters, their size and how they are embedded in space, particularly interesting for high-dimensional input spaces where visualisation is not possible.On the other hand, we will study spatially continuous datasets assuming as standing on an underlying 1D structure that we aim to learn. To this end, we resort to a regularisation of the Gaussian Mixture Model in which a spatial graph is used as a prior to approximate the underlying 1D structure. The overall graph is efficiently learnt by means of the Expectation-Maximisation algorithm with guaranteed convergence and comes together with the learning of the local width of the structure. We then illustrate applications of the algorithm to model and identify the filamentary pattern drawn by the galaxy distribution of the Universe in cosmological datasets.
  

September

• Friday, 17th of September, 11h, in room 445 "Patio", building 650 and also online: Aurélien Decelle (Theoretical physics lab of Universidad Complutense de Madrid): Equilibrium and non-Equilibrium regimes in the learning of Restricted Boltzmann Machines → recording
  Abstract

  Training Restricted Boltzmann Machines (RBMs) has been challenging for a long time due to the difficulty of computing precisely the log-likelihood gradient. Over the past decades, many works have proposed more or less successful training recipes but without studying the crucial quantity of the problem: the mixing time, i.e. the number of Monte Carlo iterations needed to sample new configurations from a model. In this work, we show that this mixing time plays a crucial role in the dynamics and stability of the trained model, and that RBMs operate in two well-defined regimes, namely equilibrium and out-of-equilibrium, depending on the interplay between this mixing time of the model and the number of steps, k, used to approximate the gradient. We further show empirically that this mixing time increases with the learning, which often implies a transition from one regime to another as soon as k becomes smaller than this time.
  In particular, we show that using the popular k (persistent) contrastive divergence approaches, with k small, the dynamics of the learned model are extremely slow and often dominated by strong out-of-equilibrium effects. On the contrary, RBMs trained in equilibrium display faster dynamics, and a smooth convergence to dataset-like configurations during the sampling.
  Finally we discuss how to exploit in practice both regimes depending on the task one aims to fulfill: (i) short k can be used to generate convincing samples in short learning times, (ii) large k (or increasingly large) is needed to learn the correct equilibrium distribution of the RBM.
  

April

• Tuesday, 27th of April, 14h30, online: Nguyen Kim Thang (IBISC, Univ. Evry / Paris-Saclay): A bandit learning algorithm and applications
  Abstract

  We consider online bandit learning in which at every time step, an algorithm has to make a decision and then observes only its reward. The goal is to design efficient (polynomial-time) algorithms that achieve a total reward approximately close to that of the best fixed decision in hindsight. In this talk, we introduce a new notion of regularity, namely (lambda, mu)-concave functions and present a bandit learning algorithm that achieves a performance guarantee which is nearly optimal. We show applications of the framework to auction design, submodular maximization and on-going works on parameter tuning in sequence alignment, clustering, tree search (in integer optimization).
  
• Tuesday, 20th of April, 14h30, online: Vadim Strijov (Moscow Institute of Physics and Technology, Federal Research Center «Computer Science and Control» of the Russian Academy of Sciences): Model selection and multimodelling
  Abstract

  Four variables make a machine learning model selected: the model parameters and their distribution, the model structure and its distribution. Changes in data implies changes in the model structure. For a complex data set one has to deal with an ensemble of selected models. An ensemble is supposed to collect different models, which fit the data holistically. This talk will discuss principles of Bayesian model selection for a single model and for an ensemble.
  
• Tuesday, 13th of April, 14h30, online: Sylvain Chevallier (LISV - UVSQ - Univ. Paris-Saclay): Learning invariant representation for transfer learning: application to BCI → recording
  Abstract

  Riemannian approaches have demonstrated their robustness and accuracy to learn brain states, with very good results in BCI applications. These geometrical approaches could contribute to solving one of the next BCI challenges, that is to drastically reduce or even remove the calibration phase. While it is theoretically possible to learn invariant representations as leverage to address this issue, the reality far more difficult. The main problem is the data availability, as learning such subject-independent representations or even paradigm-independent representations requires a tremendous amount of data. While there are strong effort to endow the BCI community with standards and common infrastructure to access the EEG recording, the data is still scarce and heterogeneous. In this talk, I will describe how some tools borrowed from differential geometry and machine learning yield promising results to help building robust and subject-independent representations.
  
• [Data Science Department seminar] Thursday, 8th of April, 14h, online at DS seminars: Alexis Dubreuil (Institut de la Vision, Sorbonne Universités, CNRS, INSERM / Group for Neural Theory from Ecole Normale Supérieure): Explainable Recurrent Neural Networks for neuroscience modeling
  Abstract

  Artificial neural networks have demonstrated impressive results either in solving practical tasks or as a tool for scientific inquiries. However, their deployment in various applications has been hindered by the fact that their inner workings are often inaccessible. The presentation will focus on a methodology that allows to reverse-engineer recurrent neural networks (RNN), a generic class of networks particularly well suited to deal with temporally varying inputs. We will first introduce the dynamical system approach to neural computations on which this methodology relies. Then by applying it to RNN trained on neuroscience tasks, we will show how explainability allows to contribute to brain modeling. If time allows we will illustrate how this can be used to study natural language processing tasks.
  
• Tuesday, 6th of April, 14h30, online: Lotfi Chaari (IRIT, INP Toulouse): Signal et image: des problèmes inverses à l’apprentissage automatique → recording
  Abstract

  Dans cet exposé, un ensemble de travaux de recherche sera présenté sur
  le traitement des signaux et des images, avec différentes applications
  et outils méthodologiques. Ces travaux représentent le résultat de
  recherche de plus d'une dizaine d'année, et montrent l'évolution des
  approches développées en allant des problèmes inverses vers
  l'apprentissage automatique. Au passage, des outils d'optimisation
  variationnelle, bayésienne et hybride sont présentés et mis au service
  de la résolution des problèmes inverses ainsi que la mise en place de
  modèle d'apprentissage profond.
  

March

• Tuesday, 30th of March, 14h30, online: Matthieu Kowalski (L2S, Paris-Saclay): Inverse problems: from sparse time-frequency synthesis to dictionary learning" → recording
  Abstract

  We will take a tour of existing techniques to tackle linear inverse problems, mainly by exploiting sparsity and structures of the synthesis coefficients in a time-frequency dictionary. In particular, we will present how to play on the thresholding rules in the iterative thresholding algorithms used to "inverse" the problem.
  We will then see two dictionary learning techniques: one relying non-negative matrix factorization to identify the repetitive structures appearing in the time-frequency plane, the other by unrolling the iterative thresholding algorithms in order to exploit the techniques of auto-differentiation used in deep learning.
  These approaches will be illustrated on classical inverse problems in audio and images.
  
• Tuesday, 23rd of March, 10h30, online: Daniel Berrar (Tokyo Institute of Technology): High-dimensional inference and optimization, Continual learning, and Model evaluation and selection → recording
  Abstract

  In this seminar, I will give an overview of my research activities in (i) high-dimensional inference and optimization, (ii) continual learning, and (iii) model evaluation and selection. My presentation will consist of three parts. In the first part, I will present some recent results of a collaborative project in systems biology. The goal of this project was to investigate the effect of the drug 5-AzaC on the fecundity and transcriptome of the blood fluke Schistosoma mansoni. This parasite causes the neglected tropical disease schistosomiasis, and novel therapeutic drug targets are urgently needed. The second part will focus on continual learning with self-organizing incremental neural networks for both unsupervised and supervised learning. The third part will focus on practical model evaluation and selection in machine learning. The statistical comparison of machine learning algorithms is frequently underpinned by null hypothesis significance testing, and the p-value has become deeply entrenched in the current evaluation practice. I will discuss underrated problems that significance testing entails for machine learning. I will also present a new "machine learner's paradox", discuss its wider implications for machine learning, and then present a Bayesian solution. Finally, I will sketch the trajectory of my future research plans.
  
• Tuesday, 9th of March, 14h30, online: Abdourrahmane Atto (Université Savoie Mont Blanc (USMB) - LISTIC): Mesures de Performances et Mécanismes d'Attention par Apprentissage de Pénalités en Apprentissage Profond → recording
• Tuesday, 2nd of March, 14h30, online: Yaël Frégier (LML, Université d'Artois / Max Planck Institute for Mathematics, Bonn): Mind2Mind: Transfer learning for GANs → recording
  Abstract

  In this talk, we will present a new approach to the problem of transfer learning for GANs. It allows training deep neural networks with limited computational resources in the specific context of generative models. We prove rigorously, within the framework of optimal transport, a theorem that ensures the convergence of the learning of the transferred Wasserstein GAN. It is joint work with Jean-Baptiste Gouray.
  

February

• Tuesday, 23rd of February, 14h30, online: Michael Vaccaro (TAU/CentraleSupelec): AutoDL Self-Service → recording, slides
  Abstract

  Several challenges in machine learning aiming at pushing the research forward in domains such as automated machine learning have been organized in the last few years. The AutoDL challenge series (https://autodl.chalearn.org/) occured in 2019-2020. The purpose was to produce an Automated Deep Learning system which could solve any classification tasks (binary, multiclass or multilabel for any type of data: image, video, time series, text, tabular) without any human intervention.
  
  In this presentation, I will present the "AutoDL Self-Service", a project to make AutoDL solutions available to the public on the Codalab platform through an "inverted competition", to which I devoted a part of my internship in the TAU team this year. I will take advantage of this opportunity to outline the workflow of the solution of the winning team of the AutoDL competition (DeepWisdom), which is available for use behind the AutoDL Self-Service, and to analyse the path taken by their solution to win this competition. Finally, I will use a benchmark of the solutions which reached the final phase of AutoDL realized on 66 datasets to compare its performance with that of the others.
  
• Tuesday, 9th of February, 14h30, online: Riad Akrour (Intelligent Autonomous Systems group, TU Darmstadt): Entropy Regularization in RL through Interpolation → recording, slides
  Abstract

  One of the main challenges in reinforcement learning (RL) is the inapplicability of the i.i.d. data assumption. Instead, RL has to interweave data gathering with policy updates. In the approximate setting, when policy improvement cannot be guaranteed uniformly over the state space, updating policies gracefully becomes key to ensure reliability and efficiency of RL. In this talk, I will summarize three contributions using entropy regularization to effectively realize such graceful updates. In addition to the algorithmic work, I will discuss the theoretical underpinnings of each contribution in the context of RL and convex optimization. The culminating contribution will be a new convex optimization approach using interpolation-based projections which, by virtue of its simplicity, appears to perform well for non-convex RL problems too. I will conclude the talk by presenting preliminary results in explainable RL, applying the aforementioned optimization tools to obtain easier to interpret policies than 'black-box' neural networks.
  
• Friday, 5th of February (whole day): Journée Statistique et Informatique pour la Science des données à Paris Saclay→ recordings
  Details

  Programme et inscription: https://indico.math.cnrs.fr/event/6351/
  

December

• Tuesday, 15th of December, 14h30, online: Jonathan Raiman (NVIDIA / TAU): DeepType 2: Superhuman entity linking; skip data cleaning, all your need is type interactions → recording
  Abstract

  An important problem in machine learning systems is the reliance on human standardized and cleaned data to train properly. This is particularly true for tasks that require bridging knowledge bases of structured data with the unstructured real world data such as entity linking. In this talk I will describe how DeepType? 2 attains a new state of the art at entity linking on standard benchmarks (TAC KBP 2010, AIDA YAGO), outperforming all previous approaches by eliminating preprocessing complexity by making learning more end to end. In the current state of the art, DeepType? 1, type systems have been shown to be a powerful representation for learning to disambiguate entities, however this representation relies on data cleaning and rules to train, and the system cannot train on disambiguation directly but rather uses type prediction as a proxy. In this work I show that the rules and data cleaning steps can be obviated by switching to an architecture that supports using graph neural networks to handle inconsistencies in the input data and an objective function that rewards disambiguation accuracy directly. I also establish a human benchmark on entity linking by collecting human expert responses to all the documents in AIDA YAGO and TAC KBP 2010. I will share my experimental protocol, the preliminary results of this study, and discuss the implications for reaching superhuman performance on entity linking.
  
• Tuesday, 8th of December, 14h30, online: [Journal Club] Victor Berger (TAU): Presentation of the article Bayesian Deep Learning and a Probabilistic Perspective of Generalization by Andrew Gordon Wilson, Pavel Izmailov → recording
  Abstract

  The key distinguishing property of a Bayesian approach is marginalization, rather than using a single setting of weights. Bayesian marginalization can particularly improve the accuracy and calibration of modern deep neural networks, which are typically underspecified by the data, and can represent many compelling but different solutions. We show that deep ensembles provide an effective mechanism for approximate Bayesian marginalization, and propose a related approach that further improves the predictive distribution by marginalizing within basins of attraction, without significant overhead. We also investigate the prior over functions implied by a vague distribution over neural network weights, explaining the generalization properties of such models from a probabilistic perspective. From this perspective, we explain results that have been presented as mysterious and distinct to neural network generalization, such as the ability to fit images with random labels, and show that these results can be reproduced with Gaussian processes. We also show that Bayesian model averaging alleviates double descent, resulting in monotonic performance improvements with increased flexibility. Finally, we provide a Bayesian perspective on tempering for calibrating predictive distributions.
  
  https://arxiv.org/abs/2002.08791
  

November

• Tuesday, 24th of November, 15h, online: Ievgen Redko (Data Intelligence team at Hubert Curien laboratory, University Jean Monnet of Saint-Etienne): Deep Neural Networks Are Congestion Games: From Loss Landscape to Wardrop Equilibrium and Beyond → recording
  Abstract

  The theoretical analysis of deep neural networks (DNN) is arguably among the most challenging research directions in machine learning (ML) right now, as it requires from scientists to lay novel statistical learning foundations to explain their behaviour in practice. While some success has been achieved recently in this endeavour, the question on whether DNNs can be analyzed using the tools from other scientific fields outside the ML community has not received the attention it may well have deserved. In this paper, we explore the interplay between DNNs and game theory (GT), and show how one can benefit from the classic readily available results from the latter when analyzing the former. In particular, we consider the widely studied class of congestion games, and illustrate their intrinsic relatedness to both linear and non-linear DNNs and to the properties of their loss surface. Beyond retrieving the state-of-the-art results from the literature, we argue that our work provides a very promising novel tool for analyzing the DNNs and support this claim by proposing concrete open problems that can advance significantly our understanding of DNNs when solved.
  
• Monday, 2nd of November, 14h30, online: Pierre Jobic (TAU/BioInfo): Demography Inference with deep learning on sets with attention mechanisms in population genetics → recording
  Abstract

  Demography inference from population genetics data has been well dominated by the famous likelihood-free inference algorithm: Approximate Bayesian Computation (ABC). However, recent deep learning architectures relying only on raw genomic SNP data show promising inference of effective population sizes at specific time steps. I will present our new architecture: MixAttSPIDNA. It is based on the team's previous architecture SPIDNA2 which is permutation invariant, and it adds attention mechanism that improved the features expressivity. We will analyze its performances and compare it with previous methods. MixAttSPIDNA has better performances than SPIDNA and ABC.
  

October

• Tuesday, 20th of October : CANCELLED
• Thursday, 15th of October, 14h30, online: Giancarlo Fissore (TAU): Relative gradient optimization of the Jacobian term in unsupervised deep learning → recording
  Abstract

  Learning expressive probabilistic models correctly describing the data is a ubiquitous problem in machine learning. A popular approach for solving it is mapping the observations into a representation space with a simple joint distribution, which can typically be written as a product of its marginals — thus drawing a connection with the field of nonlinear independent component analysis. Deep density models have been widely used for this task, but their likelihood-based training requires estimating the log-determinant of the Jacobian and is computationally expensive, thus imposing a trade-off between computation and expressive power. In this work, we propose a new approach for exact likelihood-based training of such neural networks. Based on relative gradients, we exploit the matrix structure of neural network parameters to compute updates efficiently even in high-dimensional spaces; the computational cost of the training is quadratic in the input size, in contrast with the cubic scaling of the naive approaches. This allows fast training with objective functions involving the log-determinant of the Jacobian without imposing constraints on its structure, in stark contrast to normalizing flows. An implementation of our method can be found at https://github.com/fissoreg/relative-gradient-jacobian
  
• Tuesday, 13th of October, 14h30, online: Ahmed Skander Karkar (Criteo): A Principle of Least Action for the Training of Neural Networks → recording, slides
  Abstract

  Neural networks have been achieving high generalization performance on many tasks despite being highly over-parameterized. Since classical statistical learning theory struggles to explain this behavior, much effort has recently been focused on uncovering the mechanisms behind it, in the hope of developing a more adequate theoretical framework and having a better control over the trained models. In this work, we adopt an alternate perspective, viewing the neural network as a dynamical system displacing input particles over time. We conduct a series of experiments and, by analyzing the network's behavior through its displacements, we show the presence of a low kinetic energy displacement bias in the transport map of the network, and link this bias with generalization performance. From this observation, we reformulate the learning problem as follows: finding neural networks which solve the task while transporting the data as efficiently as possible. This offers a novel formulation of the learning problem which allows us to provide regularity results for the solution network, based on Optimal Transport theory. From a practical viewpoint, this allows us to propose a new learning algorithm, which automatically adapts to the complexity of the given task, and leads to networks with a high generalization ability even in low data regimes.
  
• Tuesday, 6th of October, 14h30, online: Pierre Wolinski (University of Oxford): Initializing a neural network on the edge of chaos → slides
  Abstract

  How to initialize correctly weights and biases of an infinitely wide and deep neural network ? Glorot and Bengio (2010), then He et al. (2015), have proposed a simple answer based on the preservation of the variance of the preactivations during the forward pass. Afterwards, Poole et al. proposed the concept of "Edge of Chaos" in the paper "Exponential expressivity in deep neural networks through transient chaos" (2016). They proposed another definition of "correct" initialization. Instead of looking at the variance of the preactivations, they considered the evolution of the correlation between two inputs during the forward pass. This new point of view led to finer results, as the evidence of a phase-transition-like phenomenon according to the initialization distribution. Moreover, we are now able to predict the typical depth at which information can be propagated or backpropagated at initialization. Since the theoretical results of Edge of Chaos rely on an infinite-width assumption, some links have been drawn with the Neural Tangents Kernels (NTK).
  

...

April

• Monday, 6th of April, 10h30 (online): Abdourrahmane Atto (LISTIC – Université Savoie Mont Blanc): Apprentissage statistique, explicabilité et généralisabilité

March

• Friday, 13th of March, 14h (Shannon amphitheatre): Jean-Christophe Loiseau (DynFluid lab, ENSAM): Dimensionality reduction and system identification of physical systems : chaotic convection, a case study
• Friday, 6th of March, 16h (Shannon amphitheatre): Pierre Wolinski's PhD defense
• Wednesday, 3rd of March, 10h: Sarra Houiddi and Dominique Fourer (IBISC, Université d'Évry Val d'Essonne): Home Electrical Appliances Recognition using Relevant Features and Deep Neural Networks

February

• Friday, 28th of February, 11h: Rémi Flamary (Univ. Côte d'Azur): Optimal transport: Gromov-Wasserstein divergence and extensions
• Friday, 28th of February, 15h: [FormalDeep] Julien Girard (TAU/CEA-list) will present the paper Beyond the Single Neuron Convex Barrier for Neural Network Certification
• Friday, 14th of February, 11h: Stéphane Rivaud (Sony): Perceptual GAN for audio synthesis

January

• Friday, 17th of January, 11h: Amélie Héliou (Critéo): A journey to causal advertising
• Friday, 17th of January, 14h30: Loris Felardos presents the paper Biologically inspired alternatives to backpropagation through time for learning in recurrent neural nets

December

• Thursday, 19th of December, 17h (amphi Shannon): Lisheng Sun-Hosoya's PhD defense: Meta-Learning as a Markov Decision Process
• Tuesday, 17th of December, 17h (amphi Shannon): Diviyan Kalainathan's PhD defense: Generative Neural Networks to Infer Causal Mechanisms: Algorithms and Applications

November

• Friday, 29th of November, 11h: Luigi Gresele (Max Planck Institute for Intelligent Systems and Biological Cybernetics, Tübingen): The Incomplete Rosetta Stone Problem: Multi-View Nonlinear ICA, with applications to neuroimaging
• Friday, 29th of November, 14h (salle des thèses, bâtiment 650): Guillaume Doquet's PhD defense: Agnostic Feature Selection
• Friday, 22nd of November, 11h: Guillaume Charpiat (TAU): Input similarity from the neural network perspective
• Friday, 15th of November, 11h: Balthazar Donon (TAU): Graph Neural Solvers for Power Systems
• Tuesday, 12th of November, 14h30: Jonathan Raiman (OpenAI / TAU): DeepType: résolution référentielle d’entités multilingues par l’évolution de systèmes de types neuronaux
• Friday, 8th of November, 11h: Mandar Chandorkar (TAU / Centrum Wiskunde & Informatica (CWI), Amsterdam): Dynamic Time Lag Regression: Predicting What & When

October

• Friday, 11th of October, 11h: Signe Riemer-Sørensen (SINTEF digital, Norway): Machine learning in the real world
• Monday, 7th of October, 15h (Amphi Shannon): Corentin Tallec (TAU)'s PhD defense
• Thursday, 3rd of October, 14h30: Lisheng Sun (TAU): Meta-learning as a Markov Decision Process (MDP)
• Tuesday, 1st of October, 14h: Jakob Runge (German Aerospace Center, Institute of Data Science, Jena): Perspectives for causal inference on time series in Earth system sciences

September

• Thursday, 12th of September, 14h30: Victor Berger (TAU): From abstract items to latent space to observed data and back: Compositional Variational Auto-Encoder, followed by Zhengying Liu (TAU): Overview and unifying conceptualization of Automated Machine Learning & AutoCV Challenges Analysis
• Wednesday, 11th of September: DataIA day on Safety & AI, Turing building (INRIA Saclay); at 11h45 - 12h15: Julien Girard (CEA-list/TAU): Building Specifications for Perception Systems: Formal Proofs of Deep Networks Trained with Simulators
• Tuesday, 10th of September, 14h: Guillaume Doquet (TAU): Agnostic Feature Selection, followed by Pierre Wolinski (TAU): Learning with Random Learning Rates
• Tuesday, 3rd of September, 14h30: Luca Veyrin-Forrer (TAU): Learning To Run A Power Network

July

• Tuesday, 2nd of July, 14h30: Reda Alami (Orange/LRI): Memory bandits for decision-making in dynamic environments. Application to 5G optimization.

June

• Tuesday, 25th of June, 15h15: Victor Berger (TAU): Ensemblist Variational AutoEncoder: latent representation of semi-structured data, and Zhengying Liu (TAU): AutoCV Challenge Design and Baseline Results
• Tuesday, 18th of June, 14h30: Guillaume Doquet (TAU): Agnostic Feature Selection/ Sélection d'attributs agnostique
• Tuesday, 11th of June, 14h30: Ada Altieri (LPT, ENS): Introduction to the Thouless-Anderson-Palmer formalism and recent applications
• Tuesday 4th to Friday 7th of June (at ENS Cachan): summer school / workshop on Machine Learning & Formal Methods, details here

May

• Tuesday, 28th of May, 14h30: Talk canceled
• Wednesday 15th of May, 14h30: Erol Gelenbe (Imperial College): Réseaux Neuronaux Aléatoires - Solutions en Forme Produit, Apprentissage et Apprentissage Profond, Applications
• Tuesday, 14th of May, 14h30: Laurent Daudet (LightOn / Paris Diderot): Optical random features for large-scale machine learning
• Tuesday 7th of May, 14h30: Thibault Groueix & Pierre Alain Langlois(Imagine, ENPC): Deep Learning for 3D - Toward surface generation
• [DataIA] Thursday 2nd of May, 14h (Nano-Innov): Freddy Lecue (Chief AI Scientist @Thales Canada / INRIA Wimmics team): XAI - The story so far

April

• Tuesday, 23rd of April, 14h30: Julien Hay & Bich-Liên Doan (CentraleSupelec/LRI): Personnalisation de la recommandation d’articles d’actualité
• Thursday, 18th of April, 14h (salle des thèses 435, bâtiment 650): Big data, IA, sélection des données: causalités, corrélations, conséquences
  • Diviyan Kalainathan: Causalité observationnelle, découverte de liens de cause à effet sans expériences randomisées
  • Paola Tubaro: Sélectionné.e par une IA ? Algorithmes, inégalités, et les « humains dans la boucle »
• Tuesday, 16th of April, 14h30: Michele Alessandro Bucci (LIMSI): Control of chaotic dynamical system with Deep Reinforcement Learning approach

March

• Tuesday, 26th of March, 14h30 (usual room R2014): Saumya Jetley (University of Oxford): DeepInsight - An examination of the class decision functions learned by deep nets
• Tuesday, 12th of March, 14h30 (usual room R2014): Alexis Dubreuil (Group for Neural Theory, ENS): Reverse-engineering of low-rank recurrent neural networks
• Tuesday, 5th of March, 14h30 (usual room R2014): Gwendoline de Bie (ENSAE/TAU): Stochastic Deep Networks

February

• Tuesday, 26th of February, 14h30 (usual room R2014): Jean Barbier (ENS Paris & ICTP Trieste): Phase transitions in high-dimensional estimation and learning
• Wednesday, 20th of February, 11h30 (amphithéâtre Sophie Germain, Turing building, INRIA Saclay): [DataIA] Jérémie Mary (Critéo / Univ. Lille) Online advertising and strategic bidding
• Tuesday, 19th of February, 14h30 (usual room R2014): Loris Felardos (TAU team / IBPC): An Introduction to Graph Neural Networks
• For information: Friday, 15th of February: GT PASADENA seminar day
• Wednesday, 13th of February, 9h (Shannon amphitheatre, 660 building): Benjamin DONNOT's PhD defense: Deep Learning Methods for Predicting Flows in Power Grids: Novel Architectures and Algorithms

January

• For information: Wednesday, 30th of January, at IHES (full day): Statistics/Learning at Paris-Saclay
• Wednesday, 16th of January, 14h30 (usual room R2014): Corentin Tallec & Léonard Blier (TAU): T.B.A.
• Tuesday, 15th of January, 14h30 (amphi Shannon): Laurent Basara (TAU): The TrackML challenge: concept, methods and approaches
• Monday, 7th of January, 10h30 (usual room R2014): Jonathan Raiman (OpenAI): OpenAI Five: Atteindre un niveau professionnel à Dota en jouant contre soi-même

December

• Friday, 14th of December, 14h30 (usual room R2014): [ GT DeepNet ] Edouard Oyallon (CentraleSupelec): The shallow learning quest
• Thursday, 13th of December, 11h11 (usual room R2014): Julien Girard (TAU/CEA-list): A short introduction to formal methods and their applications for Robust Deep networks

November

• Thursday, 22nd of November, 11h11 (usual room R2014): Adrian Alan Pol (CERN): Machine Learning applications to CMS Data Quality Monitoring
• Thursday, 15th of November, 11h11 (usual room R2014): Philippe Esling (IRCAM): Artificial creative intelligence: variational inference and deep learning for modeling musical creativity slides

October

• Wednesday, 17th of October, 14h30 (Shannon amphitheatre): Pan Zhang (Institute of Theoretical Physics, Chinese Academy of Sciences): Solving Statistical Mechanics using Variational Autoregressive Networks
• Friday, 5th of October, 11h30 (usual room R2014): Thomas Lucas (Toth team, INRIA Grenoble): Mixed batches and symmetric discriminators for GAN training

September

• Thursday, 6th of September, 14h30 (usual room R2014): Mo Yang (TAU/CDS-LAL)'s end of internship: Prediction of storm trajectories

June

• Friday, 29th of June, 16h (Shannon amphitheatre): Thomas Schmitt (TAU)'s PhD defense: Appariements Collaboratifs des Offres et Demandes d'Emploi
• Thursday, 28th of June, 14h30 (Shannon amphitheatre): Alexandre Aussem (LIRIS - Lyon): Identifying irreducible disjoint factors in multivariate probability distributions: Application to multilabel learning
• Friday, 22nd of June, 11h (Shannon amphitheatre): Peter Bosman (CWI, Delft): Gene-pool Optimal Mixing Evolutionary Algorithms - From Foundations to Applications
• Friday, 22nd of June, 8h30 - 17h (room 1046): Isabelle Guyon's group seminar day: MEDI-CHAL / L2RPN
• June, Tuesday 12th (Shannon amphitheatre): Bérénice Huquet, Amandine Pierrot, Georges Hébrail (EDF Lab Paris-Saclay): Non-Intrusive Load Monitoring (NILM) problems and studies at EDF R&D
• June, Thursday 7th (17h): PhD seminar: Zhengying Liu: No Free Lunch Theorems
• June, Tuesday 5th: Martin Toth (TAU/CentraleSupelec): Deep Learning for skin disease diagnosis assistance

May

• May, Thursday 31st (Shannon Amphitheatre, 14h30): François Gonard's PhD defense: Cold-start recommendation: from algorithm portfolios to job applicant matching
• May, Wednesday 30th: Diviyan Kalainathan: Tutorial on Docker
• May, Tuesday 29th: Yufei Han (Symantec Research labs): Multi-label Learning with Highly Incomplete Data via Collaborative Embedding
• May, Thursday 24th: Stuart Russell (UC Berkeley): Provably Beneficial Artificial Intelligence, at the DATAIA Institute (Turing building, 11am)
• May, Monday 7th: Jean-Noël Vittaut (Paris 8): General Game Playing pour les jeux à information parfaite ou imparfaite
• May, Friday 4th, 11h: Dominique Fourer (IRCAM): Analysis of non-stationary and multicomponent signals with applications to music information retrieval

April

• April, Wednesday 25th: Joon Kwon (CMAP): Mirror descent strategies for regret minimization and approachability
• April, Tuesday 17th: Bertrand Thirion (Parietal team, Neurospin, INRIA/CEA): Statistical inference for high-dimensional data & application to brain imaging
• April, Tuesday 10th: Berna Bakir Batu (TAU team): A Reinforcement Learning Approach for Simulating Cascading Failures in Power Grids
• April, Tuesday 3rd: Benjamin Donnot (TAU team): Fast Power system security analysis with Guided Dropout

March

• March, Tuesday 27th: Nizam Makdoud (TAU team): Intrinsic Motivation, Exploration and Deep Reinforcement Learning
• March, Tuesday 20th: Hugo Richard (Parietal/TAU teams, INRIA): Data based analysis of visual cortex using deep features of videos (more information...)
• March, Tuesday 13th: David Rousseau (Laboratoire de l'Accélérateur Linéaire (LAL), Orsay): TrackML : The High Energy Physics Tracking Challenge (more information...)
• March, Tuesday 6th: Ulisse Ferrari (Institut de la Vision): Neuroscience & big-data: Collective behavior in neuronal ensembles (more information...)
• March, Friday 2nd: François Landes (IPhT): Physicists using and playing with Machine Learning tools: two examples (more information...)

February

• February, Tuesday 27th: Wendy Mackay (INRIA/LRI ExSitu team): Human-Computer Partnerships: Leveraging machine learning to empower human users (more information...)
• February, Tuesday 20th: Jérémie Sublime (ISEP): Unsupervised learning for multi-source applications and satellite image processing (more information...)
• February, Friday 16th: Rémi Leblond (INRIA Sierra team): SeaRNN: training RNNs with global-local losses (more information...)
• February, Tuesday 13th: Zoltan Szabo (CMAP & DSI, École Polytechnique): Linear-time Divergence Measures with Applications in Hypothesis Testing (more information...)

January

• January, Tuesday 23rd (usual room 2014): Olivier Goudet & Diviyan Kalainathan (TAU): End-to-end Causal Generative Neural Networks (more information...)
• January, Friday 19th, whole day (IHES): workshop stats maths/info du plateau de Saclay (more information...)
• January, Tuesday 9th (room 435, "salle des thèses", building 650): Michèle Sébag & Marc Schoenauer (TAU): Stochastic Gradient Descent: Going As Fast As Possible But Not Faster (more information...)

December

• December, Tuesday 19th, 14:30 (room 455, building 650): Antonio Vergari (LACAM, University of Bari 'Aldo Moro', Italy): Learning and Exploiting Deep Tractable Probabilistic Models (more information...)
• December, Wednesday 13th, 14:30 (room 445, building 650): Robin Girard (Mines ParisTech Sophia-Antipolis): Data mining and optimisation challenges for the energy transition (more information...)
• December, first week: break (NIPS)

November

• November, Wednesday 22th, 14:30 (room 2014): Marylou Gabrié (ENS Paris, Laboratoire de Physique Statistique): Mean-Field Framework for Unsupervised Learning with Boltzmann Machines (more information...)
• November, Friday 17th, 11:00 (Shannon amphitheatre): [ GT DeepNet ] Levent Sagun (IPHT Saclay): Over-Parametrization in Deep Learning (more information...)
• November, Wednesday 15th, 14:30 (room 2014): Diviyan Kalainathan & Olivier Goudet (TAU): Causal Generative Neural Networks (more information...)
• November, Thursday 9th, 11:00 (Shannon amphitheatre): Claire Monteleoni (CNRS-LAL / George Washington University): Machine Learning Algorithms for Climate Informatics, Sustainability, and Social Good (more information...)

October

• October, Tuesday 24th, 14:30 (Shannon amphitheatre): Benjamin Guedj (MODAL team, Inria Lille): A quasi-Bayesian perspective to NMF: theory and applications (more information...)
• October, Wednesday 18th, 14:30 (room 2014): Théophile Sanchez (TAU): End-to-end Deep Learning Approach for Demographic History Inference (more information...)
• October, Wednesday 11th, 14:00 (room 2014): Victor Estrade (TAU): Robust Deep Learning : A case study (more information...)
• October, Wednesday 4th, 14:30 (room 2014): Hugo Richard (Parietal/TAU): Data based alignment of brain fmri images (more information...)

September

• September, Tuesday 19th, 11:00 (Shannon amphitheatre): Carlo Lucibello (Politecnico di Torino): Probing the energy landscape of Artificial Neural Networks (more information...)

July

• July, Tuesday 4th, from 11:00 to 13:00 (Shannon amphitheatre): presentation of Brice Bathellier's team + MLspike by Thomas Deneux (more information...)

June

• June, Friday 30th, 14:30 (room 2014): internships presentation by Giancarlo Fissore: Learning dynamics of Restricted Boltzmann Machines, and by Clément Leroy: Free Energy Landscape in a Restricted Boltzmann Machine (RBM) (more information...)
• June, Thursday 29th, 14:30 (Shannon amphitheatre): [ GT DeepNet ] Alexandre Barachant: Information Geometry: A framework for manipulation and classification of neural time series (more information...)
• June, Tuesday 27th, 14:30 (room 2014) Réda Alami et Raphaël Féraud (Orange Labs): Memory Bandits : A bayesian Approach for the Switching Bandit Problem (more information...)
• June, Monday 12th, 14:30 (Shannon amphitheatre): [ GT DeepNet ] Romain Couillet (Centrale-Supélec): A Random Matrix Framework for BigData Machine Learning (more information...)

May

• May, Wednesday 24th, 16:00 (room 2014): Priyanka Mandikal (TAU): Anatomy Localization in Medical Images using Neural Networks (more information...)

April

• April, Friday 28th, 14:30 (Shannon amphitheatre): [ GT DeepNet ] Jascha Sohl-dickstein (Google Brain): Deep Unsupervised Learning using Nonequilibrium Thermodynamics (more information...)
• April, Tuesday 3rd: Thomas Schmitt: RecSys challenge 2017 (more information...)

March

• March, Thursday 2nd, 14:30 (Shannon amphitheatre): Marta Soare (Aalto University): Sequential Decision Making in Linear Bandit Setting (more information...)

February

• February 22nd, 11h: Enrico Camporeale (CWI): Machine learning for Space-Weather forecasting
• February, Thursday 16th (Shannon amphi.), 14h30: [ GT DeepNet ] Corentin Tallec: Unbiased Online Recurrent Optimization (more information...)
• February 14th (Shannon amphi.), 14h: [ GT DeepNet ] Victor Berger (Thales Services, ThereSIS): VAE/GAN as a generative model (more information...)

January

• January 25th, 10h30: Romain Julliard (Muséum National d'Histoire Naturelle): 65 Millions d'Observateurs (more information...)
• January 24th: Daniela Pamplona (Biovision team, INRIA Sophia-Antipolis / TAO): Data Based Approaches in Retinal Models and Analysis (more information...)

November

• November 30th: Martin Riedmiller (Google DeepMind). Deep Reinforcement learning for learning machines (more information...)
• November 29th: Amaury Habrard (Universite Jean Monnet de Saint-Etienne). Domain Adaptation with Optimal Transport: Mapping Estimation and Theory (more information...)
• November 24th: [ GT DeepNet ] Rico Sennrich (University of Edinburgh). Neural Machine Translation: Breaking the Performance Plateau (more information...)

June

• June 28th: Lenka Zdeborova (CEA,Ipht). Solvable models of unsupervised feature learning LRI_matrix_fact.pdf

Mai

• May 3rd: Emile Contal (ENS-Cachan). The geometry of Gaussian processes and Bayesian optimization. slides_semstat16.pdf

April

• April 26: Marc Bellemare (Google DeepMind). Eight Years of Research with the Atari 2600 (more information...)
• April 12: Mikael Kuusela (EPFL). Shape-constrained uncertainty quantification in unfolding elementary particle spectra at the Large Hadron Collider.(more information...)

March

• March 22nd: Matthieu Geist (Supélec Metz): Reductions from inverse reinforcement learning to supervised learning (more information...)
• March 15: Richard Wilkinson (University of Sheffield): Using surrogate models to accelerate parameter estimation for complex simulators (more information...)
• March 1st: Pascal Germain (Université Laval, Québec): A Representation Learning Approach for Domain Adaptation (more information...)

February

• February 9th: François Dufour (INRIA Bordeaux) (more information...)

January

• January 26th: Laurent Massoulié: Models of collective inference.(more information...).
• January 19th: Sébastien Gadat: Regret bounds for Narendra-Shapiro bandit algorithms (more information...)..

December

• December 15th: Joon Kwon: SPARSE REGRET MINIMIZATION.(more information...).

November

• November 19th: Phillipe Sampaio: A derivative-free trust-funnel method for constrained nonlinear optimization (more information...).

October

• October 27: Audrey Durand: Bandits for healthcare (more information...).

• October 20th: Jean Lafond: Low Rank Matrix Completion with Exponential Family Noise (more information...).

• October 13th
  • Flora Jay:Inferring past and present demography from genetic data (more information...).
  • Marcus Gallagher: Engineering Features for the Analysis and Comparison Black-box Optimization Problems and Algorithms (more information...).

September

• Sept. 28th
  • Olivier Pietquin, Approximate Dynamic Programming for Two-Player Zero-Sum Markov Games OlivierPietquin_ICML15.pdf
  • Francois Laviolette, Domain Adaptation (slides soon)

July

• July 2nd:Alaa Saade:MaCBetH : Matrix Completion with the Bethe Hessian(more information...)

June

• June 15th: Claire Monteleoni:Climate Informatics: Recent Advances and Challenge Problems for Machine Learning in Climate Science
• June 2nd: Robyn Francon: Reversing Operators for Semantic Backpropagation

May

• May 18th:Andras Gyorgy:Adaptive Monte Carlo via Bandit Allocation

April

• April 28th:Vianney Perchet:Optimal Sample Size in Multi-Phase Learning(more information...)
• April 27th:Hédi Soula, TBA
• April 21th: Gregory Grefenstette, INRIA Saclay: Personal semantics(more information...)
• April 7th: Paul Honeine: Relever deux défis majeurs en apprentissage par méthodes à noyaux:problème de pré-image et apprentissage en ligne (more information...)

March

• March 31th: Bruno Scherrer (Inria Nancy): Non-Stationary Modified Policy Iteration (more information...)
• March 24th: Christophe Schülke(ESPCI): Community detection with modularity: a statistical physics approach (more information...)
• March 10th: Balazs Kegl: Rapid Analytics and Model Prototyping (more information...)

February

• February 24th: Madalina Drugan (Vrije Universiteit Brussel, Belgium): Multi-objective multi-armed bandits (more information...)
• February 20th: Holger Hoos (University of British Columbia, Canada): séminaire MSR - see the slides
• February 17th :Aurélien Bellet: The Frank-Wolfe Algorithm: Recent Results and Applications to High-Dimensional Similarity Learning and Distributed Optimization more information...
• February 10th, Manuel Lopes 15interlearnteach.pdf

January

• January 27th :Raphaël Baillyra: Tensor factorization for multi-relational learning ((more information...)
• January 13th : Francesco Caltagirone: On convergence of Approximate Message Passing (talk_Caltagirone.pdf)
• January 6th : Emilie Kaufmann: Bayesian and frequentist strategies for sequential resource allocation (Emilie_Kauffman.pdf)

November

• November 4th :Joaquin Vanschoren:OpenML: Networked science in machine learning

October

• Oct. 28th,
  • Antoine Bureau, "Bellmanian Bandit Network"

This paper presents a new reinforcement learning (RL) algorithm called Bellmanian Bandit Network (BBN), where action selection in each state is formalized as a multi-armed bandit problem. The first contribution lies in the definition of an exploratory reward inspired from the intrinsic motivation criterion from -1-, combined with the RL reward. The second contribution is to use a network of multi-armed bandits to achieve the convergence toward the optimal Q-value function. The BBN algorithm is comparatively validated to -1-.
References:
-1- Manuel Lopes, Tobias Lang, Marc Toussaint, and Pierre-Yves Oudeyer. Exploration in model-based reinforcement learning by empirically estimating learning progress. In Neural Information Processing System (NIPS), 2012.

• Basile Mayeur

Abstract:
Taking inspiration from inverse reinforcement learning, the proposed Direct Value Learning for Reinforcement Learning (DIVA) approach uses light priors to gener- ate inappropriate behavior’s, and use the corresponding state sequences to directly learn a value function. When the transition model is known, this value function directly defines a (nearly) optimal controller. Otherwise, the value function is extended to the (state,action) space using off-policy learning.
The experimental validation of DIVA on the Mountain car shows the robustness of the approach comparatively to SARSA, based on the assumption that the tar- get state is known. Lighter assumptions are considered in the Bicycle problem, showing the robustness of DIVA in a model-free setting.

• Thomas Schmitt, "Exploration / exploitation: a free energy-based criterion"

We investigate a new strategy, based on a free energy criterion to solve the exploration/exploitation dilemma. Our strategy promotes exploration using an entropy term.

• Oct. 14th, Holger Hoos Slides attached.

September

• Sept. 29th, Rich Caruana

Old seminars