6.1.1 R packages

• OrdinalLBM: https://cran.r-project.org/web/packages/ordinalLBM/index.html
• FunLBM: https://cran.r-project.org/web/packages/funLBM/index.html
• SpinyReg: https://cran.r-project.org/web/packages/spinyReg/index.html

6.1.2 Python

• HDMI: https://gist.github.com/ahoudard/677943abbfc7308d30fbbe749d353d68
• MIWAE: https://github.com/pamattei/miwae
• not-MIWAE: https://github.com/nbip/notMIWAE

6.1.3 Julia

• PEN: https://github.com/SamuelWiqvist/PENs-and-ABC

6.2.1 Linkage: a statistical AI algorithm to analyze communication networks

Platform address: https://linkage.fr

Participants: Charles Bouveyron, Marco Corneli

Keywords: generative models, clustering of networks, communication networks, publication networks, cyberdefense

Collaborations: Pierre Latouche (Univ. de Paris)

Even though the clustering field has received strong attention, the joint analysis of texts and networks has been very limited while most social networks are based on texts (emails, Facebook, Twitter, ...). The implemented methodology used in the software Linkage tackles this limit and brings a solution so that texts and networks are analyzed simultaneously. Linkage implements a statistical model along with an AI algorithm for the inference which enables the segmentation of the nodes (individuals) of a network with textual edges, while identifying the topics of discussion. The networks can be directed (emails, tweets, ...) or undirected (scientific papers, posts on Facebook, ...). Linkage only requires the data of a set of textual exchanges between people or more generally between entities. For instance, we may consider the exchanges of texts between individuals on a social network, or the exchanges of emails between employees from the same company, or even the co-publications of patents or scientific papers. Linkage is also able to automatically determine the number of groups of individuals and the number of topics used. Thus, data processing is completely automatic. The outputs of the software are the classification of individuals in groups (as well as probabilities) and a description of topics with the most specific words. A demonstration platform is available at the following address: https://linkage.fr. It allows anyone wishing to, to use Linkage with their own data or with public data. Registration is free on the website but the volume of data to be treated by the platform is limited. The best computing libraries and the newest security protocols are used to guarantee an optimal treatment of data. The platform offers different ways of uploading data. First, it is possible to upload personal data in a simple format (.csv), but also to collect personal email networks via our Gmail App and finally, it is possible to run public queries (on Twitter, PubMed, Arxiv, ...). There are many graphic layouts available on the platform to facilitate the use of the results. To finish, if needed, raw data can be downloaded to run them afterwards in a visualization software such as Gephi for example.

6.2.2 Topix: a AI-based solution allowing to summarize massive and sparse text data

Platform address: https://topix.mi.parisdescartes.fr

Participants: Charles Bouveyron, Marco Corneli

Keywords: generative models, model-based co-clustering, massive and sparse text data

Collaborations: Pierre Latouche (Univ. de Paris), Laurent Bergé (Univ. Luxembourg)

Topix is an innovative AI-based solution allowing to summarize massive and possibly extremely sparse data bases involving text. Topix is a versatile technology that can be applied in a large variety of situations where large matrices of texts / comments / reviews are written by users on products or addressed to other individuals (bi-partite networks). The typical use case consists in a e-commerce company interested in understanding the relationship between its users and the sold products thanks to the analysis of user comments. A simultaneous clustering (co-clustering) of users and products is produced by the Topix software, based on the key emerging topics from the reviews and by the underlying model. The Topix demonstration platform allows to upload the user data on the website, in a totally secured framework, and let the AI-based software analyze them for the user. The platform also proposes some typical use cases to give a better idea of what Topix can do.

6.2.3 DiagnoseNET: Automatic Framework to Scale Neural Networks on Heterogeneous Systems

Platform address: https://diagnosenet.github.io/

Participants: John Anderson Garcia Henao, Michel Riveill

Keywords: federated learning, low power deep learning

Collaborations: Felix Armendo Meija (Univ. Industrial Santander, Colombia) and Calos Jaime Barrios Hernandezt Bergé (Univ. Industrial Santander, Colombia)

DiagnoseNET (video on YouTube: https://www.youtube.com/watch?v=qM11jmYTmrs) 23 is designed as a modular framework that enables the deep learning application-workflow management and expressivity to build and finetune the neural architecture, while its runtime abstracts the distributed orchestration of portability and scalability from a GPU workstation to multi-nodes computational platforms. It automatizes in one expression API the neural architecture definition, the hyperparameter search, the data locality and batching, while the runtime coordinate the parameters between the devices according to the execution modes, which enables the workers selection through synchronous or asynchronous coordination gradient computations with gRPC or MPI communication protocols, tested on x86 and arm architectures.

7.1.1 Dynamic Co-Clustering

Participants: Charles Bouveyron, Marco Corneli, Giulia Marchello

Keywords: generative models, dynamic co-clustering, count data, pharmacovigilance

Collaborations: CHU de Nice (Centre de Pharmacovigilance)

We considered in 29, 30 the problem of co-clustering count matrices that may evolved along the time and we introduce a generative model to handle it. The proposed model, named dynamic latent block model, extend the classical latent block model to the dynamic case. The modeling of the dynamic in a continuous time relies on a non- homogeneous Poisson process, with a latent partition of time intervals. We proposed to use the SEM-Gibbs algorithm for model inference. An application to pharmacovigilance is currently under consideration.

7.1.2 A Bayesian approach for clustering and exact finite-sample model selection in longitudinal data mixtures

Participants: Marco Corneli, Elena Erosheva

Keywords: longitudinal data, model based clustering, model selection, ICL, BIC, Gaussian processes

In 46, we consider mixtures of longitudinal trajectories, where one trajectory contains measurements over time of the variable of interest for one individual and each individual belongs to one cluster. The number of clusters as well as individual cluster memberships are unknown and must be inferred. We propose an original Bayesian clustering framework that allows us to obtain an exact finite-sample model selection criterion. Our approach is more flexible and parsimonious than asymptotic alternatives such as Bayesian Information Criterion (BIC) or Integrated Classification Likelihood (ICL) criterion in the choice of the number of clusters. Moreover, our approach has other desirable qualities: i) it keeps the computational effort of the clustering algorithm under control and ii) it generalizes to several families of regression mixture models, from linear to purely non-parametric.

7.1.3 Bayesian discriminative Gaussian clustering

Participants: Charles Bouveyron

Keywords: generative models, model-based clustering, Bayesian modeling, high-dimensional data,

Collaborations: Pierre Latouche (Univ. de Paris), Nicolas Jouvin (Univ. Paris 1 & Institut Curie)

High-dimensional data clustering has become and remains a challenging task for modern statistics and machine learning, with a wide range of applications. We considered in 53 the powerful discriminative latent mixture model, and we extended it to the Bayesian framework. Modeling data as a mixture of Gaussians in a low-dimensional discriminative subspace, a Gaussian prior distribution is introduced over the latent group means and a family of twelve submodels are derived considering different covariance structures. Model inference is done with a variational EM algorithm, while the discriminative subspace is estimated via a Fisher-step maximizing an unsupervised Fisher criterion. An empirical Bayes procedure is proposed for the estimation of the prior hyper-parameters, and an integrated classification likelihood criterion is derived for selecting both the number of clusters and the submodel. The performances of the resulting Bayesian Fisher-EM algorithm are investigated in two thorough simulated scenarios, regarding both dimensionality as well as noise and assessing its superiority with respect to state-of-the-art Gaussian subspace clustering models. In addition to standard real data benchmarks, an application to single image denoising is proposed, displaying relevant results. This work comes with a reference implementation for the R software in the FisherEM package accompanying the paper.

7.1.4 A New Class of EM Algorithms. Escaping Local Maxima and Handling Intractable Sampling

Participants: Juliette Chevallier

Keywords: mixture models, variational inference

Collaborations: V. Debavelaere (CMAP), Stéphanie Allassonnière (Univ. de Paris)

The expectation-maximization (EM) algorithm is a powerful computational technique for maximum likelihood estimation in incomplete data models. When the expectation step cannot be performed in closed form, a stochastic approximation of EM (SAEM) can be used. The convergence of the SAEM toward critical points of the observed likelihood has been proved and its numerical efficiency has been demonstrated. However, sampling from the posterior distribution may be intractable or have a high computational cost. Moreover, despite appealing features, the limit position of this algorithm can strongly depend on its starting one. To cope with this two issues, we propose in 5 new stochastic approximation version of the EM in which we do not sample from the exact distribution in the expectation phase of the procedure. We first prove the convergence of this algorithm toward critical points of the observed likelihood. Then, we propose an instantiation of this general procedure to favor convergence toward global maxima. Experiments on synthetic and real data highlight the performance of this algorithm in comparison to the SAEM and the EM when feasible.

7.1.5 Co-Clustering of Multivariate Functional Data for Air Pollution Analysis

Participants: Charles Bouveyron

Keywords: generative models, model-based co-clustering, functional data, air pollution, public health

Collaborations: J. Jacques and A. Schmutz (Univ. de Lyon), Fanny Simoes and Silvia Bottini (MDlab, MIS, Univ. Côte d'Azur)

In 44, we focused on Air pollution, which is nowadays a major treat for public health, with clear links with many diseases, especially cardiovascular ones. The spatio-temporal study of pollution is of great interest for governments and local authorities when deciding for public alerts or new city policies against pollution raise. The aim of this work is to study spatio-temporal profiles of environmental data collected in the south of France (Région Sud) by the public agency AtmoSud. The idea is to better understand the exposition to pollutants of inhabitants on a large territory with important differences in term of geography and urbanism. The data gather the recording of daily measurements of five environmental variables, namely three pollutants (PM10, NO2, O3) and two meteorological factors (pressure and temperature) over six years. Those data can be seen as multivariate functional data: quantitative entities evolving along time, for which there is a growing need of methods to summarize and understand them. For this purpose, a novel co-clustering model for multivariate functional data is defined. The model is based on a functional latent block model which assumes for each co-cluster a probabilistic distribution for multivariate functional principal component scores. A Stochastic EM algorithm, embedding a Gibbs sampler, is proposed for model inference, as well as a model selection criteria for choosing the number of co-clusters. The application of the proposed co-clustering algorithm on environmental data of the Région Sud allowed to divide the region composed by 357 zones in six macro-areas with common exposure to pollution. We showed that pollution profiles vary accordingly to the seasons and the patterns are conserved during the 6 years studied. These results can be used by local authorities to develop specific programs to reduce pollution at the macro-area level and to identify specific periods of the year with high pollution peaks in order to set up specific prevention programs for health. Overall, the proposed co-clustering approach is a powerful resource to analyse multivariate functional data in order to identify intrinsic data structure and summarize variables profiles over long periods of time.

7.1.6 Semi-supervised Consensus Clustering Based on Frequent Closed Itemsets

Participants: Frédéric PRECIOSO (Université Côte d'Azur)

Keywords: Clustering; Semi-supervised learning; Semi-supervised consensus clustering; Frequent closed itemsets

Collaborations: Tianshu YANG (Université Côte d'Azur, Amadeus), Nicolas PASQUIER (Université Côte d'Azur), Antoine HOM (Amadeus), Laurent DOLLE (Amadeus), in a CIFRE PhD project with Amadeus

Semi-supervised consensus clustering integrates supervised information into consensus clustering in order to improve the quality of clustering. In this paper, we study the novel Semi-MultiCons semi-supervised consensus clustering method extending the previous MultiCons approach 60. Semi-MultiCons aims to improve the clustering result by integrating pairwise constraints in the consensus creation process and infer the number of clusters K using frequent closed itemsets extracted from the ensemble members. Experimental results show that the proposed method outperforms other state-of-art semi-supervised consensus algorithms.

7.1.7 Hierarchical clustering with discrete latent variable models and the ICL criterion

Participants: Charles Bouveyron

Keywords: generative models, model-based clustering, model selection, discrete latent variable models, networks

Collaborations: Pierre Latouche (Univ. de Paris), Nicolas Jouvin (Univ. Paris 1 & Institut Curie), E. Côme (Univ. Gustave Eiffel)

In 45, we introduce a two step methodology to extract a hierarchical clustering. This methodology considers the integrated classification likelihood criterion as an objective function, and applies to any discrete latent variable models (DLVM) where this quantity is tractable. The first step of the methodology involves maximizing the criterion with respect to the discrete latent variables state with uninformative priors. To that end we propose a new hybrid algorithm based on greedy local searches as well as a genetic algorithm which allows the joint inference of the number $K$ of clusters and of the clusters themselves. The second step of the methodology is based on a bottom-up greedy procedure to extract a hierarchy of clusters from this natural partition. In a Bayesian context, this is achieved by considering the Dirichlet cluster proportion prior parameter $\alpha$ as a regularisation term controlling the granularity of the clustering. This second step allows the exploration of the clustering at coarser scales and the ordering of the clusters an important output for the visual representations of the clustering results. The clustering results obtained with the proposed approach, on simulated as well as real settings, are compared with existing strategies and are shown to be particularly relevant. This work is implemented in the R package greed.

7.1.8 Ensemble Clustering Based Semi-Supervised Learning for Revenue Accounting Workflow Management

Participants: Frédéric PRECIOSO (Université Côte d'Azur)

Keywords: Consensus Clustering; Closed sets; Anomalies correction; Multi-level clustering; Ensemble clustering; Revenue accounting workflow

Collaborations: Tianshu YANG (Université Côte d'Azur, Amadeus), Nicolas PASQUIER (Université Côte d'Azur), in a CIFRE PhD project with Amadeus

We present a semi-supervised ensemble clustering framework for identifying relevant multi-level clusters, regarding application objectives, in large datasets and mapping them to application classes for predicting the class of new instances 21. This framework extends the MultiCons closed sets based multiple consensus clustering approach but can easily be adapted to other ensemble clustering approaches. It was developed to optimize the Amadeus S.A.S revenue accounting workflow management. Revenue accounting in travel industry is a complex task when travels include several transportations, with associated services, performed by distinct operators and on geographical areas with different taxes and currencies for example. Preliminary results show the relevance of the proposed approach for the automation of workflow anomaly corrections.

7.1.9 Clustering of count data through a mixture of multinomial PCA

Participants: Charles Bouveyron

Keywords: generative models, model-based clustering, count data, medicine

Collaborations: Pierre Latouche (Univ. de Paris), Nicolas Jouvin (Univ. Paris 1 & Institut Curie), Guillaume Bataillon and Alain Livartowski (Institut Curie)

In 10, we consider count data which are becoming more and more ubiquitous in a wide range of applications, with datasets growing both in size and in dimension. In this context, an increasing amount of work is dedicated to the construction of statistical models directly accounting for the discrete nature of the data. Moreover, it has been shown that integrating dimension reduction to clustering can drastically improve performance and stability. In this paper, we rely on the mixture of multinomial PCA, a mixture model for the clustering of count data, also known as the probabilistic clustering-projection model in the literature. Related to the latent Dirichlet allocation model, it offers the flexibility of topic modeling while being able to assign each observation to a unique cluster. We introduce a greedy clustering algorithm, where inference and clustering are jointly done by mixing a classification variational expectation maximization algorithm, with a branch and bound like strategy on a variational lower bound. An integrated classification likelihood criterion is derived for model selection, and a thorough study with numerical experiments is proposed to assess both the performance and robustness of the method. Finally, we illustrate the qualitative interest of the latter in a real-world application, for the clustering of anatomopathological medical reports, in partnership with expert practitioners from the Institut Curie hospital.

7.1.10 Clustering multivariate functional data in group-specic functional subspaces

Participants: Charles Bouveyron

Keywords: generative models, model-based clustering, multivariate functional data, air pollution

Collaborations: J. Jacques and A. Schmutz (Univ. de Lyon)

With the emergence of numerical sensors in many aspects of every- day life, there is an increasing need in analyzing multivariate functional data. In 13, we focused on the clustering of such functional data, in order to ease their modeling and understanding. To this end, a novel clustering technique for multivariate functional data is presented. This method is based on a functional latent mixture model which fits the data in group-specific functional subspaces through a multivariate functional principal component analysis. A family of parsimonious models is obtained by constraining model parameters within and between groups. An EM algorithm is proposed for model inference and the choice of hyper-parameters is addressed through model selection. Numerical experiments on simulated datasets highlight the good performance of the proposed methodology compared to existing works. This algorithm is then applied to the analysis of the pollution in French cities for one year.

7.1.11 Exact Bayesian model selection for principal component analysis

Participants: Charles Bouveyron, Pierre-Alexandre Mattei

Keywords: Model selection, generative models, Bayesian modeling, high-dimensional data

Collaborations: Pierre Latouche (Univ. de Paris)

Principal component analysis (PCA) is the most commonly used method to reduce the dimensionality of a data set. By linearly transforming $p$ features into $d$ principal components, PCA somehow allows to bypass the curse of dimensionality. A key question pertaining the use of PCA is what number $d$ of principal components should we choose? The framework of Bayesian model selection, that we recently reviewed in 57, provides, in theory, a satisfying answer. Indeed, by relying on a probabilisitic model for PCA, it allows to find the dimensionality with largest posterior probability. However, there are two main practical obstacles that prevent to perform Bayesian model selection for PCA: (i) the fact that marginal likelihood of Bayesian PCA models are very difficult to compute, and (ii) the question of the choice of a relevant prior distribution. In this work 6, we address both issues and propose a simple way to perform exact model selection for Bayesian PCA. To this end, we introduce a novel Bayesian formulation of PCA based on a normal‐gamma prior distribution. In this context, we exhibit a closed‐form expression of the marginal likelihood which allows to infer an optimal number of components. This allows to address (i) in a simple way: we can compute the marginal likelihood exactly, rather than having to rely on approximations or expensive Monte Carlo sampling. To address (ii), we propose a heuristic based on the expected shape of the marginal likelihood curve in order to choose the hyperparameters. In nonasymptotic frameworks, we show on simulated data that this exact dimensionality selection approach is competitive with both Bayesian and frequentist state‐of‐the‐art methods.

7.1.12 International Patent: Clustering Techniques for Revenue Accounting Error-Handling Automation

Participants: Frédéric PRECIOSO (Université Côte d'Azur)

Keywords: Revenue accounting workflow; Semi-supervised consensus clustering; Frequent closed itemsets;

Collaborations: Tianshu YANG (Université Côte d'Azur, Amadeus), Nicolas PASQUIER (Université Côte d'Azur), Antoine HOM (Amadeus), Laurent DOLLE (Amadeus) in a CIFRE PhD project with Amadeus

This invention proposed by Amadeus SAS and Université Cote d’Azur (UCA) concerns the application of machine learning techniques for the revenue accounting task management 61. The revenue accounting system automatically process documents until an error occurs. The process is then interrupted and set to manual for validation or error correction, resulting in additional costs and delays. This invention aims to decrease cost time and manual efforts that Amadeus and its customers spend on tasks 20. This goal will be achieved by: 1) automatic identification of anomaly patterns through the unsupervised clustering of error tickets to form clusters of tickets corresponding to similar anomalies and requiring similar correction processes 2) Automatic or semi-automatic, depending on the type of the anomaly pattern, validation and/or correction of the error by the analysis of logs of correction actions taken by the correctors, associated to each cluster of tickets for the automation of the error correction process.

7.2.1 Theoretical properties of Importance Weighted variational inference

Participants: Pierre-Alexandre Mattei

Keywords: deep learning, neural networks, generative models, variational inference

Collaborations: Jes Frellsen (Technical University of Denmark)

Importance weighted variational inference (IWVI) is a promising strategy for learning latent variable models. IWVI uses new variational bounds, known as Monte Carlo objectives (MCOs), obtained by replacing intractable integrals by Monte Carlo estimates—usually simply obtained via importance sampling. These bounds ${ℒ}_K\left(Q\right)$ essentially depend on two things: the number of samples used $K$, and the distribution of the importance weights $Q$. It was previously shown that increasing the number of importance samples $L$ provably tightens the gap between the bound and the likelihood. Inspired by this simple monotonicity theorem, we present a series of nonasymptotic results that link properties of Monte Carlo estimates to tightness of MCOs 31. We challenge in particular the rationale that smaller Monte Carlo variance leads to better bounds. Moreover, we show that increasing the negative dependence of importance weights monotonically increases the bound. To this end, we use the supermodular order $⪯$ as a measure of dependence. Roughly speaking, for two random vectors $𝐰\sim Q_1$ and $𝐯\sim Q_2$, $Q_1⪯Q_2$ means that the coordinates of $𝐰\sim Q_1$ are more negatively dependent that those of $𝐯\sim Q_2$). We then show that negative dependence tightens the bound in the following sense : $Q_1⪯Q_2$ implies ${ℒ}_K\left(Q_1\right)\ge {ℒ}_K\left(Q_2\right)$ Our simple result provides theoretical support to several different approaches that leveraged negative dependence to perform efficient variational inference of deep generative models.

7.2.2 T-CAM: class activation maps for text classification

Participants: Frederic Precioso, Marco Corneli, Laurent Vanni

Keywords: Text classification, Convolutional Neural Networks, Recurrent Neural Networks, Class activation map

Collaborations: Damon Mayaffre, CNRS

The main aim of the works 43, 40 is to bring interpretability of deep networks for text classification to another level. Although some popular feature extraction techniques (e.g.attention in RNNs or Text Deconvolution Saliency for CNNs) can detect relevant linguistic markers used by a neural net to achieve classification, the same techniques are unable to assess how these markers are used. In other words, they cannot indicate whether a relevant linguistic marker provides support for or against aclass assignment. In order to overcome this limitation, we generalize the Class Activation Map (CAM) technique from computer vision to textual feature extraction. The new text CAM (T-CAM) simultaneously assesses the positive/negative contribution of any relevant input feature with respect to any class and it can be computed for several architectures (including CNNs and RNNs). Finally, T-CAM requires no sampling, neither binarization in multiclass classification problems, thus making it an attractive alternative to LIME. This work has led to several other contributions in the field of linguistics 33, 32.

7.2.3 Explaining the explainer: a first theoretical analysis of LIME

Participants: Damien Garreau

Keywords: machine learning, interpretability, statistical learning theory

Collaborations: Ulrike von Luxburg (Tübingen University)

Machine learning is used more and more often for sensitive applications, sometimes replacing humans in critical decision-making processes. As such, interpretability of these algorithms is a pressing need. One popular algorithm to provide interpretability is LIME (Local Interpretable Model-Agnostic Explanation). In 24, we provide the first theoretical analysis of LIME. We derive closed-form expressions for the coefficients of the interpretable model when the function to explain is linear. The good news is that these coefficients are proportional to the gradient of the function to explain: LIME indeed discovers meaningful features. However, our analysis also reveals that poor choices of parameters can lead LIME to miss important features.

7.2.4 Looking deeper into tabular LIME

Participants: Damien Garreau

Keywords: machine learning, interpretability, statistical learning theory

Collaborations: Ulrike von Luxburg (Tübingen University)

Interpretability of machine learning algorithm is a pressing need. Numerous methods appeared in recent years, but do they make sense in simple cases? In 50, we present a thorough theoretical analysis of Tabular LIME. In particular, we show that the explanations provided by Tabular LIME are close to an explicit expression in the large sample limit. We leverage this knowledge when the function to explain has some nice algebraic structure (linear, multiplicative, or depending on a subset of the coordinates) and provide some interesting insights on the explanations provided in these cases. In particular, we show that Tabular LIME provides explanations that are proportional to the coefficients of the function to explain in the linear case, and provably discards coordinates unused by the function to explain in the general case.

7.2.5 An analysis of LIME for text data

Participants: Damien Garreau

Keywords: machine learning, interpretability, statistical learning theory, natural language processing

Collaborations: Dina Mardaoui (Polytech Nice)

Text data are increasingly handled in an automated fashion by machine learning algorithms. But the models handling these data are not always well-understood due to their complexity and are more and more often referred to as “black-boxes.” Interpretability methods aim to explain how these models operate. Among them, LIME has become one of the most popular in recent years. However, it comes without theoretical guarantees: even for simple models, we are not sure that LIME behaves accurately. In 56, we provide a first theoretical analysis of LIME for text data. As a consequence of our theoretical findings, we show that LIME indeed provides meaningful explanations for simple models, namely decision trees and linear models.

7.2.6 Visualizing ECG Contribution into Convolutional Neural Network Classification

Participants: Frédéric Precioso

Keywords: machine learning, interpretability, statistical learning theory, natural language processing

Collaborations: Yaowei Li (UCA & School of Instrument Science and Engineering, Southeast University Nanjing, China), Chengyu Liu (The State Key Laboratory of Bioelectronics, School of Instrument Science and Engineering Southeast University Nanjing, China)

Convolutional Neural Network (CNN) demonstrated impressive classification performance on ElectroCardioGram (ECG) analysis and has a great potential to extract salient patterns from signal. Visualizing local contributions of ECG input to a CNN classification model can both help us understand why CNN can reach such high-level performances and serve as a basis for new diagnosis recommendations. In 42, we build a single-layer 1-D CNN model on ECG classification with a 99% accuracy 42. The trained model is then used to build a 1-D Deconvolved Neural Network (1-D DeconvNet) for visualization. We propose Feature Importance Degree Heatmap (FIDH) to interpret the contribution of each point of ECG input to CNN classification results, and thus to show which part of ECG raises attention of the classifier. We also illustrate the correlation between two visualization methods: first-order Taylor expansion and multilayer 1-D DeconvNet.

7.3.1 Cumulative learning enables convolutional neural network representations for small mass spectrometry data classification

Participants: Frédéric Precioso

Keywords: machine learning, cumulative learning, transfer learning, medical domain

Collaborations: Khawla Seddiki (Univ. Laval, Quebec, CA), Philippe Saudemont (Université de Lille, Inserm), Nina Ogrinc (Université de Lille, Inserm), Maxence Wisztorski (Université de Lille, Inserm), Michel Salzet (Université de Lille, Inserm), Isabelle Fournier (Université de Lille, Inserm), Arnaud Droit (Univ. Laval, Quebec, CA)

Rapid and accurate clinical diagnosis remains challenging. A component of diagnosis tool development is the design of effective classification models with Mass spectrometry (MS) data. Some Machine Learning approaches have been investigated but these models require time-consuming preprocessing steps to remove artifacts, making them unsuitable for rapid analysis. Convolutional Neural Networks (CNNs) have been found to perform well under such circumstances since they can learn representations from raw data. However, their effectiveness decreases when the number of available training samples is small, which is a common situation in medicine. In this work, we investigate transfer learning on 1D-CNNs, then we develop a cumulative learning method when transfer learning is not powerful enough. We propose to train the same model through several classification tasks over various small datasets to accumulate knowledge in the resulting representation 14. By using rat brain as the initial training dataset, a cumulative learning approach can have a classification accuracy exceeding 98% for 1D clinical MS-data. We show the use of cumulative learning using datasets generated in different biological contexts, on different organisms, and acquired by different instruments. Here we show a promising strategy for improving MS data classification accuracy when only small numbers of samples are available.

7.3.2 NEWMA: a new method for scalable model-free online change-point detection

Participants: Damien Garreau

Keywords: change-point detection, time series, kernel methods

Collaborations: Nicolas Keriven (GIPSA lab), Iacopo Poli (LightOn)

We consider the problem of detecting abrupt changes in the distribution of a multi-dimensional time series, with limited computing power and memory. In 11, we propose a new, simple method for model-free online change-point detection that relies only on fast and light recursive statistics, inspired by the classical Exponential Weighted Moving Average algorithm (EWMA). The proposed idea is to compute two EWMA statistics on the stream of data with different forgetting factors, and to compare them. By doing so, we show that we implicitly compare recent samples with older ones, without the need to explicitly store them. Additionally, we leverage Random Features (RFs) to efficiently use the Maximum Mean Discrepancy as a distance between distributions, furthermore exploiting recent optical hardware to compute high-dimensional RFs in near constant time. We show that our method is significantly faster than usual non-parametric methods for a given accuracy.

7.4.1 Co-Clustering of ordinal data via latent continuous random variables and Not Missing at Random Entries

Participants: Charles Bouveyron, Marco Corneli

Keywords: generative models, model-based clustering, count data, medicine

Collaborations: Pierre Latouche (Univ. de Paris)

This work 8 is about the co-clustering of ordinal data. Such data are very common on e-commerce platforms where customers rank the products/services they bought. In more detail, we focus on arrays of ordinal (possibly missing) data involving two disjoint sets of individuals/objects corresponding to the rows/columns of the arrays. Typically, an observed entry (i, j) in the array is an ordinal score assigned by the individual/row i to the object/column j. A new generative model for arrays of ordinal data is introduced along with an inference algorithm for parameters estimation. The model accounts for not missing at random data and relies on latent continuous random variables. The fitting allows to simultaneously co-cluster the rows and columns of an array. The estimation of the model parameters is performed via a classification expectation maximization algorithm. A model selection criterion is formally obtained to select the number of row and column clusters. To show that our approach reaches and often outperforms the state of the art, we carry out numerical experiments on synthetic data. Finally, applications on real datasets highlight the model capacity to deal with very sparse arrays.

7.4.2 Deep generative modelling with missing not at random data

Participants: Pierre-Alexandre Mattei

Keywords: missing data, neural networks, deep learning, generative models,

Collaborations: Jes Frellsen, Niel Bruun Ipsen (Technical University of Denmark)

When a missing process depends on the missing values themselves, it needs to be explicitly modelled and taken into account while doing likelihood-based inference. We present an approach for building and fitting deep latent variable models (DLVMs) in cases where the missing process is dependent on the missing data. Specifically, a deep neural network enables us to flexibly model the conditional distribution of the missingness pattern given the data. This allows for incorporating prior information about the type of missingness (e.g. self-censoring) into the model. Our inference technique, based on importance-weighted variational inference, involves maximising a lower bound of the joint likelihood. Stochastic gradients of the bound are obtained by using the reparameterisation trick both in latent space and data space. Our method is called not-missing-at-random importance-weighted autoencoder (not-MIWAE)17. We show on various kinds of data sets and missingness patterns that explicitly modelling the missing process can be invaluable. We apply our method to censoring in datasets from the UCI database, clipping in images and the issue of selection bias in recommender systems.

7.4.3 DeepLTRS: A Deep Latent Recommender System based on User Ratings and Reviews

Participants: Dingge Liang, Marco Corneli, Charles Bouveyron.

Keywords: Recommender System, Topic modelling, latent representation learning.

Collaborations: Pierre Latouche (Univ. de Paris)

We introduce in 28, 54 a deep latent recommender system (deepLTRS) in order to provide users with high quality recommendations based on observed user ratings and texts of product reviews. Our approach adopts a variational auto-encoder architecture as a generative deep latent variable model for both an ordinal matrix encoding users scores about products, and a document-term matrix encoding the reviews. Moreover, an alternated user/product mini-batching optimization structure is proposed to jointly capture user and product preferences. Numerical experiments on simulated and real-world data sets demonstrate that deepLTRS outperforms the state-of-the-art, in particular in contexts of extreme data sparsity.

7.4.4 Hierarchical Multimodal Attention for Deep Video Summarization

Participants: Melissa Sanabria, Frédéric Precioso, (Inria, CNRS, I3S, Maasai, Université Côte d'Azur).

Keywords: Event stream data, Soccer match data, Video Summarization, Multimodal data, Sports Analytics

Collaborations: Thomas Menguy (Wildmoka Company), this work has been co-funded by Région Sud Provence Alpes Côte d'Azur (PACA), Université Côte d'Azur (UCA) and Wildmoka Company.

This paper explores the problem of summarizing professional soccer matches as automatically as possible using both the event-stream data collected from the field and the content broadcasted on TV. We have designed an architecture, introducing first (1) a Multiple Instance Learning method that takes into account the sequential dependency among events and then (2) a hierarchical multimodal attention layer that grasps the importance of each event in an action 38. We evaluate our approach on matches from two professional European soccer leagues, showing its capability to identify the best actions for automatic summarization by comparing with real summaries made by human operators.

7.4.5 Profiling Actions for Sport Video Summarization: An attention signal analysis

Participants: Melissa Sanabria, Frédéric Precioso, (Inria, CNRS, I3S, Maasai, Université Côte d'Azur).

Keywords: Video Summarization, Sports Video, Event stream data, Neural Network, User-Centric

Collaborations: Thomas Menguy (Wildmoka Company), this work has been co-funded by Région Sud Provence Alpes Côte d'Azur (PACA), Université Côte d'Azur (UCA) and Wildmoka Company.

Currently, in broadcast companies many human operators select which actions should belong to the summary based on multiple rules they have built upon their own experience using different sources of information. These rules define the different profiles of actions of interest that help the operator to generate better customized summaries. Most of these profiles do not directly rely on broadcast video content but rather exploit metadata describing the course of the match. In 19, we show how the signals produced by the attention layer of a recurrent neural network can be seen as a learned representation of these action profiles and provide a new tool to support operators' work. The results in soccer matches show the capacity of our approach to transfer knowledge between datasets from different broadcasting companies, from different leagues, and the ability of the attention layer to learn meaningful action profiles.

7.4.6 A coherent framework for learning spatiotemporal piecewise-geodesic trajectories from longitudinal manifold-valued data

Participants: Juliette Chevallier

Keywords: Bayesian inference, differential geometry, manifold-valued data

Collaborations: V. Debavelaere (CMAP), Stéphanie Allassonnière (Univ. de Paris)

We provide in 7 a coherent framework for studying longitudinal manifold-valued data. We introduce a Bayesian mixed-effects model which allows estimating both a group-representative piecewise-geodesic trajectory in the Riemannian space of shape and inter-individual variability. We prove the existence of the maximum a posteriori estimate and its asymptotic consistency under reasonable assumptions. Due to the non-linearity of the proposed model, we use a stochastic version of the Expectation-Maximization algorithm to estimate the model parameters. Our simulations show that our model is not noise-sensitive and succeeds in explaining various paths of progression.

7.4.7 How to deal with missing data in supervised deep learning?

Participants: Pierre-Alexandre Mattei

Keywords: missing data, neural networks, deep learning, generative models,

Collaborations: Jes Frellsen, Niel Bruun Ipsen (Technical University of Denmark)

The issue of missing data in supervised learning has been largely overlooked, especially in the deep learning community. In 26, we investigate strategies to adapt neural architectures to handle missing values. Here, we focus on regression and classification problems where the features are assumed to be missing at random. Of particular interest are schemes that allow to reuse as-is a neural discriminative architecture. One scheme involves imputing the missing values with learnable constants. We propose a second novel approach that leverages recent advances in deep generative modelling. More precisely, a deep latent variable model can be learned jointly with the discriminative model, using importance-weighted variational inference in an end-to-end way. This hybrid approach, which mimics multiple imputation, also allows to impute the data, by relying on both the discriminative and the generative model. We also discuss ways of using a pre-trained generative model to train the discriminative one. In domains where powerful deep generative models are available, the hybrid approach leads to large performance gains.

7.4.8 NIH Peer Review: Criterion Scores Completely Account for Racial Disparities in Overall Impact Scores

Participants: Elena A. Erosheva

Keywords: grant funding, hierarchical models, matching, multilevel models, racial disparities

Collaborations: Sheridan Grant, Mei-Ching Chen, Mark D. Lindner, Richard K. Nakamura, Carole J. Lee

Previous research has found that (a) funding disparities are driven by applications’ final impact scores and (b) only a portion of the black/white funding gap can be explained by bibliometrics and topic choice. Using NIH R01 applications for council years 2014-2016, we show in 9 that black/white funding disparities remain. We then examine the relationship between assigned reviewers’ preliminary overall impact scores and preliminary criterion scores—Significance, Investigator, Innovation, Approach, and Environment—to evaluate whether racial disparities in preliminary overall impact scores can be explained by other application and applicant characteristics. We hypothesize that differences in commensuration—the process of combining criterion scores into overall impact scores—disadvantage black applicants. Using multilevel models and matching on key variables including career stage, gender, and area of science to reduce model dependence, we find little evidence for racial disparities emerging in the process of combining preliminary criterion scores into preliminary overall impact scores. At the same time, we find that preliminary criterion scores fully account for racial disparities—yet do not explain all of the variability—in preliminary overall impact scores. Future research should focus on understanding the extent to which disparities in preliminary criterion scores could be driven by implicit racial preferences, black-white differences in bibliometric outputs, topic choice, and/or current or cumulative experiences impacting grantsmanship.

7.4.9 Machine Learning Approaches to Epidemic Models

Participants: Marco Corneli, Marco Gori

Keywords: Epidemic models, spatiotemporal diffusion models, variational learning models

Collaborations: Andrea Zugarini (UNISI), Enrico Meloni (UNISI), Alessandro Betti (UNISI)

The COVID-19 outbreak has stimulated the interest in the proposal of novel epidemiological models to predict the course of the epidemic so as to help planning effective control strategies. In particular, in order to properly interpret the available data, it has become clear that one must go beyond most classic epidemiological models and consider richer description of the stages of infection. The problem of learning the parameters of these models is of crucial importance especially when assuming that they are time-variant, which further enriches their effectiveness. In this project 59 we propose a general approach for learning time-variant parameters of dynamic compartmental models from epidemic data. We formulate the problem in terms of a functional risk that depends on the learning variables through the solutions of a dynamic system. The resulting variational problem is then solved by using a gradient flow on a suitable, regularized functional. We forecast the epidemic evolution in Italy and France.

8.1.1 Orange

Participants: Hugo Miralles, Michel Riveill

External participants: Tamara Tosic (Orange), Thierry Nagellen (Orange)

Value: 45000 EUR

This collaboration contract is a CIFRE contract built upon the PhD of Hugo Miralles on Distributed device-embedded classification and prediction in near-to-real time. In this thesis we study the problem of efficient classification and prediction of multivariate time-series captured by embedded devices by using joint data-model distributed algorithms for applications that preserve private data.

8.1.2 Pro-BTP

Participants: Michel Riveill, Mansour Zoubeirou

External participants: Frédéric Estroumza (Pro BTP), Philippe Darnault (Pro BTP)

Value: Under negotiation

This collaboration contract is a CIFRE contract built upon the PhD of Mansour Zoubeirou on Detection of weak signals and incremental evolution of prediction models, especially in the context of fraud detection.

8.1.3 NXP

Participants: Frederic Precioso, Charles Bouveyron, Baptiste Pouthier (Ph.D. candidate)

External participants: Laurent Pilati (NXP)

Value: 45000 EUR

This collaboration contract is a CIFRE contract built upon the PhD of Baptiste Pouthier on Deep Learning and Statistical Learning on audio-visual data for embedded systems.

8.1.4 Oscaro

Participants: Charles Bouveyron, Pierre-Alexandre Mattei

External participants: Warith Harchaoui (Ph.D. candidate), Nils Grunwald (Oscaro)

Value: 45000 EUR

This collaboration contract is a CIFRE contract built upon the PhD of Warith Harchaoui on representation learning using neural networks and optimal transport.

8.1.5 Ezako

Participant: Frederic Precioso

External participants: Mireille Blay-Fornarino (Univ. Côte d'Azur), Yassine El Amraoui (Ezako - Univ. Côte d'Azur), Julien Muller (Ezako), Bora Kizil (Ezako)

Value: 45000 EUR

This collaboration contract is a CIFRE contract built upon the PhD of Yassine El Amraoui on Maximizing expert feedback in the detection of anomalies in time series.

8.1.6 Amadeus

Participant: Frederic Precioso

External participants: Nicolas Pasquier (Univ. Côte d'Azur), Tianshu Yang (Amadeus - Univ. Côte d'Azur), Antoine Hom (Amadeus), Laurent Dolle (Amadeus), Mickael Defoin-Platel (Amadeus), Luca Marchetti (Amadeus)

Value: 60000 EUR

This collaboration contract is a CIFRE contract built upon the PhD of Tianshu Yang on Semi-supervised clustering applied in revenue accounting.

8.1.7 Detection and characterization of salient moments for automatic summaries

Participants: Melissa Sanabria, Frédéric Precioso.

External Participants: Thomas Menguy (Wildmoka)

Value: 45000 EUR

Keywords: Video Summarization, Multimodal data, Soccer match data.

We have designed an architecture, introducing a Multiple Instance Learning method that takes into account the sequential dependency among events and a hierarchical multimodal attention layer that grasps the importance for each event in an action. We evaluate our approach on matches from two professional European soccer leagues, showing its capability to identify the best actions for automatic summarization by comparing with real summaries made by human operators.

8.2.1 Grant from the Novo Nordisk foundation

Participant: Pierre-Alexandre Mattei

External participants: Wouter Boomsma (University of Copenhagen), Jes Frellsen (Technical University of Denmark), Søren Hauberg (principal investigator, Technical University of Denmark), and Ole Winther (Technical University of Denmark)

Value: 180000 DKK ($\approx 24000$ EUR)

The object of this grant is the organisation of the 2nd Copenhagen Workshop on Generative Models and Uncertainty Quantification (GenU) in 2021.

9.1.1 Inria international partners

9.1.2 Participation in other international programs

SAILAB: Streaming Virtual Reality: Learning for Attentional Models and Network Optimization

Participants: Miguel Fabián Romero Rondón, Frédéric Precioso

Keywords: Modeling and prediction, Virtual Reality, Deep Learning, Multimedia Streaming

Collaborations: Lucile Sassatelli (Project Leader), Ramon Aparicio-Pardo, from I3S Lab of Université Côte d'Azur, and Dario Zanca and Marco Gori, University of Siena (to define gravitational laws to model attention in VR videos).

The goal of the project is to optimally decide what to transmit from a video sphere of VR content to maximize the streaming quality as observed in Fig. 24. We have made several contributions in 2020 37, 18.

9.2.1 Visits of international scientists

In 2020, we had the opportunity to welcome a few scientists, for several weeks each:

• Elena Erosheva: She is a full professor of Statistics at the University of Washington, Seattle (USA). She is also an International Chair in Data Science at Université Côte d'Azur, funded by the Idex Jedi program,
• Marco Gori: He is a full professor of Computer Science at Universita di Sienna, Sienna (Italy). He is also an International Chair of the Institut 3IA Côte d'Azur,
• Riccardo Rastelli: He is an assistant professor in Statistics at University College Dublin. His research interests include network analysis and Bayesian statistics.

9.3.1 FP7 & H2020 Projects

Maasai is one of the 3IA-UCA research teams of AI4Media, one of the 4 ICT-48 Center of Excellence in Artificial Intelligence which has started in September 2020. There are 30 partners (Universities and companies), and 3IA-UCA received about 325k€.

Institut 3IA Côte d'Azur

Following the call of President Macron to found several national institutes in AI, we presented in front of a international jury our project for the Institut 3IA Côte d'Azur in April 2019. The project was selected for funding (50 M€ for the first 4 years, including 16 M€ from the PIA program) and started in september 2019. Charles Bouveyron and Marco Gori are two of the 29 3IA chairs which were selected ab initio by the international jury and is also acting in 2020 as the Deputy Scientific Director of the institute. The research of the institute is organized around 4 thematic axes: Core elements of AI, Computational Medicine, AI for Biology and Smart territories. The Maasai reserch team is totally aligned with the first axis of the Institut 3IA Côte d'Azur. The team has 4 Ph.D. students who are directly funded by the institute.

Web site: https://3ia.univ-cotedazur.eu

Centre de pharmacovigilance, CHU Nice

Participants: Charles Bouveyron, Marco Corneli, Giulia Marchello, Michel Riveill, Xuchun Zhang

Keywords: Pharmacovigilance, co-clustering, count data, text data

Collaborateurs: Milou-Daniel Drici, Audrey Freysse, Fanny Serena Romani

The team works very closely with the Regional Pharmacovigilance Center of the University Hospital Center of Nice (CHU) through several projects. The first project concerns the construction of a dashboard to classify spontaneous patient and professional reports, but above all to report temporal breaks. To this end, we are studying the use of dynamic co-classification techniques to both detect significant ADR patterns and identify temporal breaks in the dynamics of the phenomenon. The second project focuses on the analysis of medical reports in order to extract, when present, the adverse events for characterization. After studying a supervised approach, we are studying techniques requiring fewer annotations.

Interpretability for automated decision services

Participants: Damien Garreau, Frédéric Precioso

Keywords: interpretability, deep learning

Collaborations: Greger Ottosson (IBM)

Businesses rely more and more frequently on machine learning to make automated decisions. In addition to the complexity of these models, a decision is rarely by using only one model. Instead, the crude reality of business decision services is that of a jungle of models, each predicting key quantities for the problem at hand, that are then agglomerated to produce the final decision, for instance by a decision tree. In collaboration with IBM, we want to provide principled methods to obtain interpretability of these automated decision processes. An intern will start on the subject in March 2021.

10.1.1 Scientific events: organisation

10.1.2 Scientific events: selection

10.1.3 Invited talks

• Charles Bouveyron was invited for a keynote lecture at the European Conference on Data Analysis, Napoli, Italy, September 2020 (the conference was cancelled due to the Covid pandemy).
• Pierre-Alexandre Mattei gave a talk at the online One World ABC seminar (video on YouTube).

10.1.4 Leadership within the scientific community

• Charles Bouveyron has been the Deputy Scientific Director of the Institut 3IA Côte d'Azur from September 2019 until December 2020.

10.1.5 Scientific expertise

Most permanent members of the team are serving as experts for the ANR or foreign research agencies.

10.1.6 Research administration

• Frédéric Precioso is the Scientific Responsible for AI at the French Research Agency (ANR) since September 2011.

10.2.1 Teaching

C. Bouveyron, D. Garreau, F. Precioso and M. Riveill are professors at Université Côte d'Azur and therefore handle usual teaching duties. M. Corneli and P.A. Mattei are also teaching around 60h per year at Université Côte d'Azur.

C. Bouveyron (up to august 2020) and M. Riveill (since September 2020) are responsible for the MSc. Data Sciences and Artificial Intelligence at Université Côte d'Azur.

10.2.2 Supervision

PhD students, postdocs, and interns of the team are listed in Section 1. Additionally, members of the team supervise several Masters projects, in particular from the MSc. Data Sciences and Artificial Intelligence at Université Côte d'Azur.

10.3.1 Interventions

• Frederic Precioso has developed an experimental platform both for research projects and scientific mediation on the topic of autonomous cars. This platform is currently installed in the "Maison de l'Intelligence Artificielle" where high school students have already experimented coding autonomous remote control cars (https://maison-intelligence-artificielle.com/experimenter-projets-ia/).
• Charles Bouveyron has developed an interactive software allowing to visualise the relationships between pollution and a health disease (dispnea) in the Région Sud. This platform is currently installed in the "Maison de l'Intelligence Artificielle".

International journals

• 5 article StéphanieS. Allassonnière and JulietteJ. Chevallier. A New Class of EM Algorithms. Escaping Local Maxima and Handling Intractable Sampling Computational Statistics and Data Analysis December 2020
  HAL back to text
• 6 article CharlesC. Bouveyron, PierreP. Latouche and Pierre-AlexandreP.-A. Mattei. Exact Dimensionality Selection for Bayesian PCA Scandinavian Journal of Statistics 2020
  HAL DOI back to text
• 7 article JulietteJ. Chevallier, VianneyV. Debavelaere and StéphanieS. Allassonnière. A coherent framework for learning spatiotemporal piecewise-geodesic trajectories from longitudinal manifold-valued data SIAM Journal on Imaging Sciences 2021
  HAL back to text
• 8 article MarcoM. Corneli, CharlesC. Bouveyron and PierreP. Latouche. Co-clustering of ordinal data via latent continuous random variables and not missing at random entries Journal of Computational and Graphical Statistics 2020
  HAL DOI back to text
• 9 articleElenaE. Erosheva, SheridanS. Grant, Mei-ChingM.-C. Chen, MarkM. Lindner, RichardR. Nakamura and CaroleC. Lee. NIH peer review: Criterion scores completely account for racial disparities in overall impact scoresScience Advances623June 2020, eaaz4868
  HALDOIback to text
• 10 article NicolasN. Jouvin, PierreP. Latouche, CharlesC. Bouveyron, GuillaumeG. Bataillon and AlainA. Livartowski. Greedy clustering of count data through a mixture of multinomial PCA Computational Statistics 2020
  HAL DOI back to text
• 11 articleNicolasN. Keriven, DamienD. Garreau and IacopoI. Poli. NEWMA: a new method for scalable model-free online change-point detectionIEEE Transactions on Signal Processing68April 2020, 3515 - 3528
  HALDOIback to text
• 12 article AlexandreA. Saint-Dizier, JulieJ. Delon and CharlesC. Bouveyron. A unified view on patch aggregation Journal of Mathematical Imaging and Vision 2020
  HAL DOI
• 13 article AmandineA. Schmutz, JulienJ. Jacques, CharlesC. Bouveyron, LaurenceL. Cheze and PaulineP. Martin. Clustering multivariate functional data in group-specific functional subspaces Computational Statistics February 2020
  HAL DOI back to text
• 14 article KhawlaK. Seddiki, PhilippeP. Saudemont, FrédéricF. Precioso, NinaN. Ogrinc, MaxenceM. Wisztorski, MichelM. Salzet, IsabelleI. Fournier and ArnaudA. Droit. Cumulative learning enables convolutional neural network representations for small mass spectrometry data classification Nature Communications 11 1 December 2020
  HAL DOI back to text
• 15 article Trang UyenT. Tran, Ha ThanhH. Ti Hoang, PhuongP. Hoai Dang and MichelM. Riveill. Multitask Aspect_Based Sentiment Analysis with Integrated Bidirectional LSTM & CNN Model IAES International Journal of Artificial Intelligence (IJ-AI) 9 1 March 2020
  HAL DOI
• 16 article DarioD. Zanca, MarcoM. Gori, StefanoS. Melacci and AlessandraA. Rufa. Gravitational Models Explain Shifts on Human Visual Attention Scientific Reports 2020
  HAL

International peer-reviewed conferences

• 17 inproceedings Niels BruunN. Ipsen, Pierre-AlexandreP.-A. Mattei and JesJ. Frellsen. not-MIWAE: Deep Generative Modelling with Missing not at Random Data International Conference on Learning Representations Virtual conference (formerly planned in Vienna), Austria https://openreview.net/forum?id=tu29GQT0JFy 2021
  HAL back to text
• 18 inproceedingsMiguel FabianM. Romero Rondon, LucileL. Sassatelli, RamonR. Aparicio-Pardo and FrédéricF. Precioso. A Unified Evaluation Framework for Head Motion Prediction Methods in 360° VideosMMSys '20 - 11th ACM Multimedia Systems Conference20MMSys '20: Proceedings of the 11th ACM Multimedia Systems ConferenceIstanbul, TurkeyJune 2020, 279–284
  HALDOIback to text
• 19 inproceedings MelissaM. Sanabria, FredericF. Precioso and ThomasT. Menguy. Profiling Actions for Sport Video Summarization: An attention signal analysis MMSP 2020 - 22nd IEEE International Workshop on Multimedia Signal Processing Tampere, Finland September 2020
  HAL back to text
• 20 inproceedingsTianshuT. Yang, NicolasN. Pasquier, AntoineA. Hom, LaurentL. Dollé and FrédéricF. Precioso. Semi-supervised Consensus Clustering Based on Frequent Closed Itemsets: Amadeus Intellectual Property Invention Patent ID2326WW00 "Clustering Techniques for Revenue Accounting Error-Handling Automation" Defensive PaperCIKM'2020 29th ACM International Conference on Information and Knowledge Management (Acceptance Rate: 18%)Proceedings of the CIKM'2020 29th ACM International Conference on Information and Knowledge ManagementGalway, Irelandhttps://www.cikm2020.org/October 2020, 3341-3344
  HALDOIback to text
• 21 inproceedingsTianshuT. Yang, NicolasN. Pasquier and FrédéricF. Precioso. Ensemble Clustering Based Semi-Supervised Learning for Revenue Accounting Workflow Management: Amadeus Intellectual Property Invention Patent ID2326WW00 "Clustering Techniques for Revenue Accounting Error-Handling Automation" Defensive PaperDATA'2020 International Conference on Data Science, Technology and Applications (Acceptance Rate: 14%)Proceedings of the DATA'2020 International Conference on Data Science, Technology and ApplicationsParis, Francehttp://www.dataconference.orgJuly 2020, pp. 283-293
  HALDOIback to text

Conferences without proceedings

• 22 inproceedingsGabrieleG. Ciravegna, FrancescoF. Giannini, MarcoM. Gori, MarcoM. Maggini and StefanoS. Melacci. Human-Driven FOL Explanations of Deep LearningIJCAI-PRICAI 2020 - 29th International Joint Conference on Artificial Intelligence and the 17th Pacific Rim International Conference on Artificial IntelligenceYokohama, JapanJuly 2020, 2234-2240
  HALDOI
• 23 inproceedings John AJ. García, FrédéricF. Precioso, PascalP. Staccini and MichelM. Riveill. DiagnoseNET: Automatic Framework to Scale Neural Networks on Heterogeneous Systems Applied to Medical Diagnosis ICITCS 2020 - 8th International Conference on IT Convergence and Security Nha Trang / Virtual, Vietnam August 2020
  HAL back to text
• 24 inproceedings DamienD. Garreau and UlrikeU. Von Luxburg. Explaining the Explainer: A First Theoretical Analysis of LIME AISTATS 2020 - 23rd International Conference on Artificial Intelligence and Statistics Palermo /Online, Italy August 2020
  HAL back to text
• 25 inproceedings WangW. Hao, ChristelC. Dartigues-Pallez and MichelM. Riveill. Supervised learning for Human Action Recognition from multiple Kinects BDMS 2020 - 7th Big Data Management and Service in DASFAA 2020 Jeju, South Korea September 2020
  HAL
• 26 inproceedings Niels BruunN. Ipsen, Pierre-AlexandreP.-A. Mattei and JesJ. Frellsen. How to deal with missing data in supervised deep learning? Artemiss - ICML Workshop on the Art of Learning with Missing Values Vienne, Austria https://artemiss-workshop.github.io/ July 2020
  HAL back to text
• 27 inproceedings Luis C.L. Lamb, ArturA. Garcez, MarcoM. Gori, MarceloM. Prates, PedroP. Avelar and MosheM. Vardi. Graph Neural Networks Meet Neural-Symbolic Computing: A Survey and Perspective IJCAI-PRICAI 2020 - 29th International Joint Conference on Artificial Intelligence - Pacific Rim International Conference on Artificial Intelligence Yokohama, Japan July 2020
  HAL
• 28 inproceedings DinggeD. Liang, MarcoM. Corneli, PierreP. Latouche and CharlesC. Bouveyron. Missing rating imputation based on product reviews via deep latent variable models ICML2020 Workshop on the Art of Learning with Missing Values (Artemiss) Virtual, France July 2020
  HAL back to text
• 29 inproceedings GiuliaG. Marchello, MarcoM. Corneli and CharlesC. Bouveyron. The Dynamic Latent Block Model for Sparse and Evolving Count Matrices ICML Workskshop Artemiss Nice, France July 2020
  HAL back to text
• 30 inproceedings GiuliaG. Marchello, MarcoM. Corneli and CharlesC. Bouveyron. The dynamic latent block model for the co-clustering of evolving binary matrices SFdS 2020 - 52èmes journées de Statistique de la la Société Française de Statistique Nice, France June 2020
  HAL back to text
• 31 inproceedings Pierre-AlexandreP.-A. Mattei and JesJ. Frellsen. Negative Dependence Tightens Variational Bounds ICML 2020 - 2nd Workshop on Negative Dependence and Submodularity for ML Vienne / Online, Austria July 2020
  HAL back to text
• 32 inproceedings DamonD. Mayaffre, CamilleC. Bouzereau, MagaliM. Guaresi, FrédéricF. Precioso and LaurentL. Vanni. Du texte à l'intertexte. Le palimpseste Macron au révélateur de l'Intelligence artificielle CMLF 2020 - 7ème Congrès mondiale de linguistique française Montpellier / Online, France July 2020
  HAL back to text
• 33 inproceedings DamonD. Mayaffre and LaurentL. Vanni. Objectiver l'intertexte ? Emmanuel Macron, deep learning et statistique textuelle JADT 2020 - 15èmes Journées Internationales d'Analyse statistique des Données Textuelles Toulouse, France June 2020
  HAL back to text
• 34 inproceedings EnricoE. Meloni, LucaL. Pasqualini, MatteoM. Tiezzi, MarcoM. Gori and StefanoS. Melacci. SAILenv: Learning in Virtual Visual Environments Made Simple 25th International Conference on Pattern Recognition Milan, Italy January 2021
  HAL
• 35 inproceedings FannyF. Orlhac, ThibautT. Cassou-Mounat, Jean-YvesJ.-Y. Pierga, MarieM. Luporsi, ChristopheC. Nioche, CharlesC. Bouveyron, NicholasN. Ayache, NinaN. Jehanno, AlainA. Livartowski and IreneI. Buvat. Can we identify "twin patients" to predict response to neoadjuvant chemotherapy in breast cancer? SNMMI Annual Meeting Virtual Meeting, United States July 2020
  HAL
• 36 inproceedings FannyF. Orlhac, Anne-CapucineA.-C. Rollet, IrèneI. Buvat, JacquesJ. Darcourt, VéroniqueV. Bourg, ChristopheC. Nioche, CharlesC. Bouveyron, NicholasN. Ayache and OlivierO. Humbert. Identifying a reliable radiomic signature from scarce data: illustration for 18F-FDOPA PET images in glioblastoma patients EANM Annual Meeting - Annual Meeting of the European Association of Nuclear Medicine Virtual Meeting, Austria October 2020
  HAL
• 37 inproceedings Miguel FabianM. Romero Rondon, LucileL. Sassatelli, RamonR. Aparicio-Pardo and FrédéricF. Precioso. TRACK: A Multi-Modal Deep Architecture for Head Motion Prediction in 360-Degree Videos IEEE International Conference on Image Processing (ICIP) Abu Dhabi, United Arab Emirates October 2020
  HAL back to text
• 38 inproceedings MelissaM. Sanabria, FrédéricF. Precioso and ThomasT. Menguy. Hierarchical Multimodal Attention for Deep Video Summarization 25th International Conference on Pattern Recognition Milan, Italy January 2021
  HAL back to text
• 39 inproceedings MatteoM. Tiezzi, StefanoS. Melacci, AlessandroA. Betti, MarcoM. Maggini and MarcoM. Gori. Focus of Attention Improves Information Transfer in Visual Features NeurIPS 2020 - 34th Conference on Neural Information Processing Systems Vancouver / Online, Canada December 2020
  HAL
• 40 inproceedings LaurentL. Vanni, MarcoM. Corneli, DominiqueD. Longrée, DamonD. Mayaffre and FrédéricF. Precioso. Hyperdeep : deep learning descriptif pour l'analyse de données textuelles JADT 2020 - 15èmes Journées Internationales d'Analyse statistique des Données Textuelles Toulouse, France June 2020
  HAL back to text

Scientific book chapters

• 41 inbookCharlesC. Bouveyron. High-Dimensional Statistical Learning and Its Application to Oncological Diagnosis by RadiomicsHealthcare and Artificial IntelligenceMarch 2020, 121-128
  HALDOI
• 42 inbookYaoweiY. Li, FrédéricF. Precioso and ChengyuC. Liu. Visualizing ECG Contribution into Convolutional Neural Network ClassificationFeature Engineering and Computational Intelligence in ECG MonitoringJune 2020, 157-174
  HALDOIback to textback to text
• 43 inbookLaurentL. Vanni, MarcoM. Corneli, DominiqueD. Longrée, DamonD. Mayaffre and FrédéricF. Precioso. Key Passages : From statistics to Deep LearningText Analytics. Advances and Challenges2020, 41-54
  HALDOIback to text

Reports & preprints

• 44 misc CharlesC. Bouveyron, JulienJ. Jacques, AmandineA. Schmutz, FannyF. Simoes and SilviaS. Bottini. Co-Clustering of Multivariate Functional Data for the Analysis of Air Pollution in the South of France June 2020
  HAL back to text
• 45 misc EtienneE. Côme, PierreP. Latouche, NicolasN. Jouvin and CharlesC. Bouveyron. Hierarchical clustering with discrete latent variable models and the integrated classification likelihood April 2020
  HAL back to text
• 46 misc MarcoM. Corneli and ElenaE. Erosheva. A Bayesian approach for clustering and exact finite-sample model selection in longitudinal data mixtures October 2020
  HAL back to text back to text
• 47 misc SergioS. Daga, ChiaraC. Fallerini, MargheritaM. Baldassarri, FrancescaF. Fava, FlorianaF. Valentino, GabriellaG. Doddato, ElisaE. Benetti, SimoneS. Furini, AnnaritaA. Giliberti, RossellaR. Tita, SaraS. Amitrano, MirellaM. Bruttini, IlariaI. Meloni, Anna MariaA. Pinto, FrancescoF. Raimondi, AlessandraA. Stella, FilippoF. Biscarini, NicolaN. Picchiotti, MarcoM. Gori, PietroP. Pinoli, StefanoS. Ceri, MaurizioM. Sanarico, FrancisF. Crawley, AlessandraA. Renieri, FrancescaF. Mari and ElisaE. Frullanti. Employing a Systematic Approach to Biobanking and Analyzing Genetic and Clinical Data for Advancing COVID-19 Research December 2020
  HAL DOI
• 48 misc LapoL. Faggi, AlessandroA. Betti, DarioD. Zanca, StefanoS. Melacci and MarcoM. Gori. Wave Propagation of Visual Stimuli in Focus of Attention June 2020
  HAL
• 49 misc MichaelM. Fop, Pierre-AlexandreP.-A. Mattei, CharlesC. Bouveyron and Thomas BrendanT. Murphy. Unobserved classes and extra variables in high-dimensional discriminant analysis February 2021
  HAL
• 50 misc DamienD. Garreau and UlrikeU. Von Luxburg. Looking Deeper into Tabular LIME September 2020
  HAL back to text
• 51 misc MarcoM. Gori. An Overview on the Web of Clinical Data September 2020
  HAL
• 52 misc MarcoM. Gori and AlessandroA. Betti. Backprop Diffusion is Biologically Plausible June 2020
  HAL
• 53 misc NicolasN. Jouvin, CharlesC. Bouveyron and PierreP. Latouche. A Bayesian Fisher-EM algorithm for discriminative Gaussian subspace clustering December 2020
  HAL back to text
• 54 misc DinggeD. Liang, MarcoM. Corneli, CharlesC. Bouveyron and PierreP. Latouche. DeepLTRS: A Deep Latent Recommender System based on User Ratings and Reviews November 2020
  HAL back to text
• 55 misc GiuliaG. Marchello, AudreyA. Fresse, MarcoM. Corneli and CharlesC. Bouveyron. Co-clustering of evolving count matrices in pharmacovigilance with the dynamic latent block model February 2021
  HAL
• 56 misc DinaD. Mardaoui and DamienD. Garreau. An Analysis of LIME for Text Data October 2020
  HAL back to text
• 57 misc Pierre-AlexandreP.-A. Mattei. A Parsimonious Tour of Bayesian Model Uncertainty December 2020
  HAL back to text
• 58 misc CédricC. Vincent-Cuaz, TitouanT. Vayer, RémiR. Flamary, MarcoM. Corneli and NicolasN. Courty. Online Graph Dictionary Learning February 2021
  HAL
• 59 misc AndreaA. Zugarini, EnricoE. Meloni, AlessandroA. Betti, AndreaA. Panizza, MarcoM. Corneli and MarcoM. Gori. An Optimal Control Approach to Learning in SIDARTHE Epidemic model December 2020
  HAL back to text

Other scientific publications

• 60 misc TianshuT. Yang, NicolasN. Pasquier and FrédéricF. Precioso. Semi-Supervised Consensus Clustering Based on Frequent Closed Itemsets Bodenseeforum, Germany April 2020
  HAL back to text

Patents

• 61 patent T. Yang, N. Pasquier, F. Precioso, A. Hom and L. Dollé. Clustering Techniques for Revenue Accounting Error-Handling Automation July 2020
  HAL back to text