7.1.1 Pl@ntNet

• Keywords:
  Plant identification, Deep learning, Citizen science
• Functional Description:
  Pl@ntNet is a participatory platform and information system dedicated to the production of botanical data through deep learning-based plant identification. It includes 3 main front-ends, an Android app (the most advanced and the most used one), an iOs app (being currently re-developed) and a web version. The main feature of the application is to return the ranked list of the most likely species providing an image or an image set of an individual plant. In addition, Pl@ntNet’s search engine returns the images of the dataset that are the most similar to the queried observation allowing interactive validation by the users. The back-office running on the server side of the platform is based on Snoop visual search engine (a software developed by ZENITH) and on NewSQL technologies for the data management. The application is distributed in more than 200 countries (30M downloads) and allows identifying about 35K plant species at present time.
• Publications:
  hal-01629195, hal-02937618, hal-03343235, hal-01182775
• Contact:
  Alexis Joly
• Participants:
  Antoine Affouard, Jean-Christophe Lombardo, Pierre Bonnet, Hervé Goëau, Mathias Chouet, Hugo Gresse, Julien Champ, Alexis Joly

7.1.2 ThePlantGame

• Keyword:
  Crowd-sourcing
• Functional Description:
  ThePlantGame is a participatory game whose purpose is the production of big taxonomic data to improve our knowledge of biodiversity. One major contribution is the active training of the users based on innovative sub-task creation and assignment processes that are adaptive to the increasing skills of the user. Thousands of players are registered and produce on average about tens new validated plant observations per day. The accuracy of the produced taxonnomic tags is very high (about 95%), which is quite impressive considering the fact that a majority of users are beginners when they start playing.
• Publication:
  hal-01629149
• Contact:
  Alexis Joly
• Participants:
  Maximilien Servajean, Alexis Joly

7.1.3 Savime

• Name:
  Simulation And Visualization IN-Memory
• Keywords:
  Data management., Distributed Data Management
• Functional Description:
  SAVIME is a multi-dimensional array DBMS for scientific applications. It supports a novel data model called TARS (Typed ARray Schema), which extends the basic array data model with typed arrays. In TARS, the support of application dependent data characteristics is provided through the definition of TAR objects, ready to be manipulated by TAR operators. This approach provides much flexibility for capturing internal data layouts through mapping functions, which makes data ingestion independent of how simulation data has been produced, thus minimizing ingestion time.
• Publication:
  lirmm-01620376
• Contact:
  Patrick Valduriez
• Participants:
  Hermano Lustosa, Fabio Porto, Patrick Valduriez
• Partner:
  LNCC - Laboratório Nacional de Computação Científica

7.1.4 OpenAlea

• Keywords:
  Bioinformatics, Biology
• Functional Description:
  OpenAlea is an open source project primarily aimed at the plant research community. It is a distributed collaborative effort to develop Python libraries and tools that address the needs of current and future works in Plant Architecture modeling. It includes modules to analyze, visualize and model the functioning and growth of plant architecture. It was formally developed in the Inria VirtualPlants team.
• Release Contributions:
  OpenAlea 2.0 adds to OpenAlea 1.0 a high-level formalism dedicated to the modeling of morphogenesis that makes it possible to use several modeling paradigms (Blackboard, L-systems, Agents, Branching processes, Cellular Automata) expressed with different languages (Python, L-Py, R, Visual Programming, ...) to analyse and simulate shapes and their development.
• Publications:
  hal-01166298, hal-00831811
• Contact:
  Christophe Pradal
• Participants:
  Christian Fournier, Christophe Godin Maury, Christophe Pradal, Frédéric Boudon, Patrick Valduriez, Esther Pacitti, Yann Guédon
• Partners:
  CIRAD, INRAE

7.1.5 Imitates

• Name:
  Indexing and mining Massive Time Series
• Keywords:
  Time Series, Indexing, Nearest Neighbors
• Functional Description:
  Time series indexing is at the center of many scientific works or business needs. The number and size of the series may well explode depending on the concerned domain. These data are still very difficult to handle and, often, a necessary step to handling them is in their indexing. Imitates is a Spark Library that implements two algorithms developed by Zenith. Both algorithms allow indexing massive amounts of time series (billions of series, several terabytes of data).
• Publication:
  lirmm-01886794
• Contact:
  Florent Masseglia
• Partners:
  New York University, Université Paris-Descartes

7.1.6 UMX

• Name:
  open-unmix
• Keywords:
  Source Separation, Audio
• Scientific Description:
  UMX implements state of the art audio/music source separation with deep neural networks (DNNs). It is intended to serve as a reference in the domain. It has been presented in two major scientific communications: An Overview of Lead and Accompaniment Separation in Music (https://hal-lirmm.ccsd.cnrs.fr/lirmm-01766781) and Music Separation with DNNs (Making it work (ISMIR 2018 Tutorial) https://sigsep.github.io/ismir2018_tutorial/index.html#/cover).
• Functional Description:
  UMX implements audio source separation with deep learning, using the Pytorch and Tensorflow frameworks. It comprises the code for both training and testing the separation networks, in a flexible manner. Pre- and post-processing around the actual deep neural nets include sophisticated specific multichannel filtering operations.
• Publication:
  lirmm-01766781
• Authors:
  Antoine Liutkus, Fabian Robert Stoter, Emmanuel Vincent
• Contact:
  Antoine Liutkus

7.1.7 TDB

• Keywords:
  Data assimilation, Big data, Data extraction
• Scientific Description:
  TDB comes as a building block for audio machine learning pipelines. It is a scraping tool that allows large scale data augmentation. Its different components allow building a large dataset of samples composed of related audio tracks, as well as the associated metadata. Each sample comprises a dynamic number of entries.
• Functional Description:

  TDB is composed of two core submodules. First, a data extraction pipeline allows scraping a provider url so as to extract large amounts of audio data. The provider is assumed to offer audio content in a freely-accessible way through a hardcoded specific structure. The software automatically downloads the data locally under a raw data format. To aggregate the raw data set, a list of item ids is used. The item ids will be requested from the provider given a url in parallel fashion. Second, a data transformation pipeline allows transforming the raw data into a dataset that is compatible with machine learning purposes. Each produced subfolder contains a set of audio files corresponding to a predefined set of sources, along with the associated metadata. A working example is provided.

  Each component has several submodules, in particular, network handling and audio transcoding. Thus, TDB can be viewed as an extract-transform-load (ETL) pipeline that enables applications such as deep learning on large amounts of audio data, assuming that an adequate data provider url is fed into the software.

• Contact:
  Antoine Liutkus
• Participants:
  Antoine Liutkus, Fabian Robert Stoter

7.1.8 UMX-PRO

• Name:
  Unmixing Platform - PRO
• Keywords:
  Audio signal processing, Source Separation, Deep learning
• Scientific Description:

  UMX-PRO is written in Python using the TensorFlow 2 framework and provides an off-the-shelf solution for music source separation (MSS). MSS consists in extracting different instrumental sounds from a mixture signal. In the scenario considered by UMX-PRO, a mixture signal is decomposed into a pre-defined set of so called targets, such as: (scenario 1) {"vocals", "bass", "drums", "guitar", "other"} or (scenario 2) {"vocals", "accompaniment"}.

  The following key design choices were made for UMX-PRO. The software revolves around the training and inference of a deep neural network (DNN), building upon the TensorFlow v2 framework. The DNN implemented in UMX-PRO is based on a BLSTM recurrent network. However, the software has been designed to be easily extended to other kinds of network architectures to allow for research and easy extensions. Given an appropriately formatted database (not part of UMX-PRO), the software trains the network. The database has to be split into train and valid subsets, each one being composed of folders called samples. All samples must contain the same set of audio files, having the same duration: one for each desired target. For instance: {vocals.wav, accompaniment.wav}. The software can handle any number of targets, provided they are all present in all samples. Since the model is trained jointly, a larger number of targets increases the GPU memory usage during training. Once the models have been trained, they can be used for separation of new mixtures through a dedicated end-to-end separation network. Interestingly, this end-to-end network comprises an optional refining step called expectation-maximization that usually improves separation quality.

• Functional Description:
  UMX-PRO implements a full audio separation deep learning pipeline in Tensorflow v2. It provides everything needed to train and use a deep learning model for separating music signals, including network architecture, data pipeline, training code, inference code as well as pre-trained weights. The software comes with full documentation, detailed comments and unit tests.
• Authors:
  Antoine Liutkus, Fabian Robert Stoter
• Contact:
  Antoine Liutkus

8.1.1 Data Provenance and Recommendation in ML Workflows

Participants: Esther Pacitti, Patrick Valduriez.

Scientific ML is multidisciplinary, heterogeneous, and affected by the physical constraints of the domain, making analyses challenging. In 33, we leverage workflow provenance techniques to build a holistic view to support the entire lifecycle of scientific ML. The experiments show that the decisions enable queries that integrate domain semantics with ML models while keeping low overhead (<1%), high scalability and high performance.

Provenance can also contribute to the interpretation of models resulting from the life cycle in deep neural networks (DNN). In 45, we propose a provenance data-based approach for the collection and analysis of configuration data in DNN, with an experimental validation with Keras and a real application which provides evidence of the flexibility and efficiency of the approach.

Data provenance can also be exploited for recommendation. In 32,we present FReeP-Feature Recommender from Preferences, a parameter value recommendation method that is designed to suggest values for workflow parameters, taking into account past user preferences. FReeP is based on ML techniques, particularly in Preference Learning. FReeP is composed of three algorithms, where two of them aim at recommending the value for one parameter at a time, and the third makes recommendations for $n$ parameters at once.

8.1.2 Reproducible Performance Optimization of Complex wokflows on the Edge-to-Cloud Continuum

Participants: Daniel Rosendo, Patrick Valduriez.

Complex workflows typically combine computing, analytics and learning. They often require a hybrid execution infrastructure with IoT devices interconnected to cloud/HPC systems. Such workflows are subject to complex constraints and requirements in terms of performance, resource usage, energy consumption and financial costs. This makes it challenging to optimize their configuration and deployment. In 46, 50, we propose a methodology to support the optimization of such workflows on the Edge-to-Cloud Continuum. We implement it as an extension of E2Clab, a previously proposed framework supporting the complete experimental cycle across the Edge-to-Cloud Continuum (see our white paper in 57). Our approach relies on a rigorous analysis of possible configurations in a controlled testbed environment to understand their behaviour and related performance trade-offs. We illustrate our methodology with our Pl@ntNet application.

8.2.1 Efficient Similarity Search in Large Time Series Databases

Participants: Oleksandra Levchenko, Boyan Kolev, Djamel Edine Yagoubi, Reza Akbarinia, Florent Masseglia, Dennis Shasha, Patrick Valduriez.

Fast and accurate similarity search is critical to performing many data mining tasks like motif discovery, classification or clustering. In 25, we present our parallel solutions, developed based on two state-of-the-art approaches iSAX and sketch, for $k$ nearest-neighbor (kNN) search in large databases of time series. We compare the two solutions based on various measures of quality and time performance, and propose a tool that uses the characteristics of application data to determine which solution to choose for that application and how to set the parameters for that solution. Our experiments show that: (i) iSAX and its derivatives perform best in both time and quality when the time series can be characterized by a few low frequency Fourier Coefficients, a regime where the iSAX pruning approach works well; (ii) iSAX performs significantly less well when high frequency Fourier Coefficients have much of the energy of the time series; (iii) A random projection approach based on sketches by contrast is more or less independent of the frequency power spectrum. The experiments show the close relationship between pruning ratio and time for exact iSAX as well as between pruning ratio and the quality of approximate iSAX. Our toolkit analyzes typical time series of an application (i) to determine optimal segment sizes for iSAX and (ii) when to use Parallel Sketches instead of iSAX. Our solutions have been implemented using Spark, evaluated over a cluster of nodes, and have been applied to both real and synthetic data.

8.2.2 Efficient Computation of Aggregations for Analyzing Large Streaming Data

Participants: Reza Akbarinia.

In stream processing systems, aggregations having the inherent property of summarizing information from data, constitute a fundamental operator to compute real-time statistics. They are typically computed over finite subsets of a stream, called sliding windows. One of the challenges faced by the sliding window aggregation (SWAG) algorithms is to incrementally compute aggregations over moving data, i.e., without recomputing the aggregation from scratch after inserting new data items or evicting old data items to/from the window. High throughput and low latency are essential requirements as stream processing systems are typically designed for real-time applications.

In 14, we propose PBA (Parallel Boundary Aggregator), a novel algorithm that computes incremental aggregations in parallel. PBA groups continuous slices into chunks, and maintains two buffers for each chunk containing, respectively, the cumulative slice aggregations (denoted as csa) and the left cumulative slice aggregations (denoted as lcs) of the chunk’s slices. Using PBA, SWAGs can be computed in constant time for both amortized and worst-case time. We also propose an approach to optimize the chunk size, which guarantees the minimum latency for PBA. We conducted extensive empirical experiments using both synthetic and real-world datasets. Our experiments show that PBA behaves very well for average and large sliding windows (e.g., with sizes higher than 1024 values) compared to the state-of-the-art algorithms. For small-size windows, the results show the superiority of the non-parallel version of PBA (denoted as SBA) that outperforms other algorithms in terms of throughput.

8.2.3 Anomaly Detection in Time Series

Participants: Heraldo Borges, Reza Akbarinia, Florent Masseglia.

In many real-world applications such as health monitoring, siesmology, aircraft engine test, etc, the data is collected in the form of time series. Anomaly detection is very important in this type of data. It refers to discovering any abnormal behavior within the data encountered in a specific time interval. For instance, cardiologists are interested in identifying anomalous parts of ECG signals to diagnose heart disorders.

In 18, we present a survey of existing techniques for anomaly detection in time series. The objective is to provide an understanding of the problem of detecting anomalies and how existing techniques are related. The paper is divided into three main parts. First, the main concepts are presented. Then, the anomaly detection task is defined. Afterward, the main approaches and strategies to solve the problem are presented.

8.2.4 Efficient kNN Search in Large Chemometrics Databases

Participants: Reza Akbarinia, Florent Masseglia.

In precision agriculture and plant breeding, the amount of data tends to increase, and is becoming more and more complex, leading to difficulties in managing and analyzing it. Optical instruments such as NIR Spectroscopy or hyperspectral imaging are gradually expanding directly in the field, increasing the amount of spectral database. Using these tools allows access to non-destructive and rapid measurements to classify new varieties according to breeding objectives. However, processing this massive amount of spectral data is challenging

In the context of genotype discrimination, we propose a method called parSketch-PLSDA 30 in order to analyze spectral data. ParSketch-PLSDA is a combination of our indexing technique parSketch and the reference method PLSDA for predicting classes from multivariate data. We evaluated ParSketch-PLSDA through extensive experimentation over a large spectra dataset generated from hyperspectral images of leaves of four different sunflower genotypes. We compared Sketch-PLSDA with the state-of-the-art PLSDA method. The prediction model obtained by PLSDA has a classification error close to 23% on average across all genotypes. Our ParSketch-PLSDA method outperforms PLSDA by significantly reducing the prediction error to 10%.

8.3.1 New Methods for Species Distribution Modeling at Large Scale

Participants: Benjamin Deneu, Christophe Botella, Alexis Joly, François Munoz.

In 3, we applied Convolutional Neural Networks (CNN) to a very large dataset of plant occurrences in France (GBIF), on a large taxonomical scale. We found that the landscape structure around location crucially contributes to improve predictive performance in particular for rare species, which open promising perspectives for biodiversity monitoring and conservation strategies. These models were computed on Jean ZAY super-computer as one of the Grand Challenge of GENCI.

In 39, we study for the first time a country-wide species distribution model based on very high resolution (1m) remote sensing images processed by a CNN. We demonstrate that this model can capture habitat information at very fine spatial scales, while providing overall better predictive performance than conventional models.

In 1, we propose a new method to correct the biases of SDMs coming from the non-uniform spatial sampling effort. The proposed method is based on estimating the variation in sampling effort over units of a spatial mesh in parallel with the environmental density of multiple species using a marked Poisson process model. It is particularly suited to the analysis of massive but highly heterogeneous presence-only data such as citizen science data.

8.3.2 AI-based Herbarium Specimens Analysis

Participants: Hervé Goëau, Alexis Joly.

In 26, herbarium specimens were scored both manually by human observers and by a mask R-CNN object detection model to (1) evaluate the concordance between ML and manually-derived phenological data and (2) determine whether ML-derived data can be used to reliably assess phenological patterns. The ML model generally underestimates the number of reproductive structures present on each specimen; however, when these counts are used to provide a quantitative estimate of the phenological stage of plants on a given sheet, the ML and manually-derived PI’s were highly concordant, demonstrating that phenological data extracted using machine learning can be used reliably to estimate the phenological stage of herbarium specimens and to detect phenological patterns.

In 21, we investigated whether digitized herbarium specimens can be used to improve the identification performance of species for which we have very few (if any) photos. Therefore, we introduced a new open dataset and a new domain adaptation method that provides significant improvement; however, the task remains highly challenging and progress is still needed.

8.3.3 Deep Learning Models for Digital Agriculture

Participants: Julien Champ, Herve Goeau, Alexis Joly, Baldwin Dumortier, Antoine Liutkus.

In 42, we propose a new deep learning architecture for plant disease recognition. Contrary to classical models based on the classification of host species-disease pairs, we propose a new conditional multi-task learning (CMTL) approach which allows the distribution of host species and disease characteristics learned simultaneously with a conditional link between them. We show that our approach can improve the performance of plant disease identification while improving the understanding of the prediction. Meanwhile, we also compose a new dataset that could serve as an important benchmark in this field.

In 37, we report our participation to the ROSE challenge evaluating new approaches for automatic weeding by agricultural robots. The solution we develop jointly with INRAE is based on the following tasks: weed detection and location based on a new instance segmentation deep learning framework, positioning of the weeding probe according to location results and electrical destruction of weeds. Discrimination results obtained between crops and weeds attained 76% accuracy and mortality rates of weeds ranged from 75% (Polygonaceae) to 13% (Brassicaceae). Survivors were nevertheless impacted with a severe limitation of their growth demonstrating the relevance of the overall approach.

Some current ongoing research in the context of a collaboration with INRAE (AI3P project) furthermore aims at predicting protein interaction and structure, in a way similar to how Transformer-based model predict text in natural language processing except that positions are not taken as integers, but as scalars in ${ℝ}^3$. The most notable application of this work would be to predict which protein should be synthesized to interact with some other target protein, which is a common problem in drug design. The study is risky but promising.

8.3.4 Evaluation of Species Identification and Prediction Algorithms

Participants: Alexis Joly, Herve Goeau, Benjamin Deneu, Titouan Lorieul, Camille Garcin.

In addition to LifeCLEF, we published a new plant identification dataset and benchmark in the context of NeurIPS 2021 conference 41. We highlight two particular features of this new dataset called Pl@ntNet-300K: (i) its strong class imbalance, and, (ii) the presence of many species visually similar. These two characteristics make Pl@ntNet-300K well suited for the evaluation of set-valued classification methods and algorithms. Therefore, we introduce two set-valued evaluation metrics (macro-average top-k and average-k accuracy) and we provide baseline results established by training deep neural networks using the cross-entropy loss.

8.3.5 Innovative services in Pl@ntNet platform

Participants: Alexis Joly, Benjamin Deneu, Jean-Christophe Lombardo, Antoine Affouard.

We first enriched Pl@ntNet public API (my.plantnet.org) with new features such as similar image search, automatic organ recognition or geo-localized species prediction (based on the results of Benjamin Deneu's PhD work 3). The API was published as a new service in the European Open Science Cloud. It currently has nearly 4900 user accounts including start up developers, IT services providers, researchers, citizen observatories, etc.

Besides, two new complementary concepts were developed to support the creation of customized and collaborative e-floras in the mobile application 15: groups and monitoring work spaces. Groups allow any user to create a private or public space on the platform . These groups are used by people to structure their collaborative activity related to a given area or taxonomic group. Over 3300 groups have already been created by e.g., professional land managers, educators, and plant enthusiasts. Monitoring work spaces allow a given stakeholder to access all Pl@ntNet observations and identification requests of a given species list in a particular area. For example, this service has been mobilized to follow the recent development of an invasive species (i.e., Hakea sericea Schrad. & J.C. Wendl.) in and around a natural reserve on the Mediterranean coast.

8.4.1 Setting the State of the Art in Music Demixing

Participants: Fabian-Robert Söter, Baldwin Dumortier, Antoine Liutkus.

8.4.2 Deep models for audio and long-range data

Participants: Antoine Liutkus, Fabian-Robert Söter, Baldwin Dumortier.

Our strategy is to go beyond our current expertise on music demixing to address new interesting topics in audio. This means leaving our comfort zone on source separation to address new exciting challenges. This year, we pursued our theoretical efforts on Transformers, and were the first to propose a new way to combine relative positional encodings with linear complexity Transformers 7.

10.1.1 Associate Teams in the framework of an Inria International Lab or in the framework of an Inria International Program

10.2.1 FP7 & H2020 projects

Institut de Convergence Agriculture numérique #DigitAg, (2017-2023), 275 Keuro.

Participants: Alexis Joly, Florent Masseglia, Esther Pacitti, Christophe Pradal, Patrick Valduriez.

#DigitAg brings together in a partnership of seventeen actors (public research and teaching organizations, transfer actors and companies) with the objective of accelerating and supporting the development of agriculture companies in France and in southern countries based on new tools, services and uses. Based in Montpellier with an office in Toulouse and Rennes and led by Irstea, #DigitAg's ambition is to become a world reference for digital agriculture. In this project, Zenith is involved in the analysis of big data from agronomy, in particular, plant phenotyping and biodiversity data sharing.

ANR PerfAnalytics (2021-2024), 100 Keuro.

Participants: Reza Akbarinia, Florent Masseglia.

PPR "Antibiorésistance": structuring tool "PROMISE" (2021-2024), 240 Keuro.

Participants: Reza Akbarinia, Florent Masseglia.

The objective of the PROMISE (PROfessional coMmunIty network on antimicrobial reSistancE) project is to build a large data warehouse for managing and anlyzing antimicrobial resitance (AMR) data. It gathers 21 existing professional networks and 42 academic partners from three sectors, human, animal, and environment. The project is based on the following transdisciplinary and cross-sectoral pillars: i) fostering synergies to improve the one-health surveillance of antibiotic consumption and AMR, ii) data sharing for improving the knowledge of professionals, iii) improving clinical research by analyzing the shared data.

ANR WeedElec (2018-2021), 106 Keuro.

Participants: Julien Champ, Hervé Goëau, Alexis Joly.

CASDAR CARPESO (2020-2022), 87 Keuro.

Participants: Julien Champ, Hervé Goëau, Alexis Joly.

10.3.1 Others

11.1.1 Scientific events: organisation

11.1.2 Scientific events: selection

11.1.3 Journal

11.1.4 Invited talks

• E. Pacitti: "Uma Perspectiva Evolutiva e Multidisciplinar do Tratamento de Dados", Seminários 2021, Instituto de Computação – UFF, Rio de Janeiro, 17 March 2021.
• P. Valduriez: "Innovation : startup strategies", 12th edition of the Marcus Evans "Innovation Strategies" conference, 20 May 2021.
• A. Joly: "Which scientific challenges beyond data collection?", keynote talk, CEMEB-NUMEV day, 30 March 2021.
• A. Joly: "Pl@ntNet, la science des données au service de la biodiversité végétale", EGC 2021 conference, 25 January 2021.

11.1.5 Leadership within the scientific community

• E. Pacitti: Member of the Steering Committee of the BDA conference.
• A. Joly: Foundator & scientific coordinator of LifeCLEF virtual: computer-assisted identification of living organisms (19 collaborators for the organization, 100s of registrants/participants, 100s publications, 1000s citations)
• A. Joly: Scientific and technical director of the Pl@ntNet platform: AI-based citizen science for plant biodiversity (3 permanent researchers, 4 engineers, 3 PhD students, 1 postdoc, tens of national and international partners, thousands of developers with an account on Pl@ntNet API, millions of end-users).
• A. Joly: Steering board of Cos4Cloud, H2020 research project (6M euros) aimed at integrating citizen science in the European Open Science Cloud (EOSC) through the co-design of innovative services.
• A. Liutkus: elected member of IEEE technical committee on audio.

11.1.6 Scientific expertise

• R. Akbarinia: Expert for STIC AmSud international program.
• A. Joly: Expert for the French National HPC grand equipment (GENCI)
• A. Liutkus: expert for ANR.

11.1.7 Research administration

• E. Pacitti: manager of Polytech' Montpellier's International Relationships for the computer science department (100 students).
• P. Valduriez: scientific manager for the Latin America zone at Inria's Direction of Foreign Relationships (DRI).
• F. Masseglia: deputy scientific director of Inria for the domain "Perception, Cognition And Interaction".

11.2.1 Teaching

Most permanent members of Zenith teach at the Licence and Master degree levels at UM2.

Esther Pacitti responsibilities on teaching (theoretical, home works, practical courses,exams) and supervision at Polytech' Montpellier UM, for engineer students:

• IG3: Database design, physical organization, 54h, level, L3, 50 students.
• IG4: Distributed Databases and NoSQL, 80h , level M1, 50 students.
• Large Scale Information Management (Iot, Recommendation Systems, Graph Databases), 27h, level M2, 20 students.
• Supervision of industrial projects
• Supervision of master internships.
• Supervision of computer science discovery projects.

Patrick Valduriez:

• Professional: Distributed Information Systems, Big Data Architectures, 75h, level M2, Capgemini Institut.

Alexis Joly:

• Univ. Montpellier: Machine Learning, 10h, level M2
• Polytech' Montpellier: Content-Based Image Retrieval, 6h, level M2.
• AgroParisTech: Deep Learning, 10h, level M1.

Antoine Liutkus

• Polytech' Montpellier: Audio Machine Learning, 1.5h, level M1.

Christophe Pradal

• Univ. Montpellier: Root System Modelling, 15h, level M2.
• Univ. Cambridge: Functional-Structural Plant Modelling, 20h, level M2.

11.2.2 Supervision

PhD & HDR:

• Defended PhD: Alena Shilova, Memory Saving Strategies for Deep Neural Network Training, Defended December 2021, Univ. Bordeaux. Advisors: Olivier Beaumont, Lionel Eyraud-Dubois, Alexis Joly.
• PhD in progress: Daniel Rosendo, Enabling HPC-Big Data Convergence for Intelligent Extreme-Scale Analytics, started Oct 2019, Univ. Rennes. Advisors: Gabriel Antoniu, Alexandru Costan, Patrick Valduriez.
• PhD in progress: Benjamin Deneu, Large-scale and High-resolution Species Distribution Modelling using Deep Learning, Univ. Montpellier. Advisors: Alexis Joly, François Munoz, Maximilien Servajean, Pierre Bonnet.
• PhD in progress: Camille Garcin, Multi-class classification with high label ambiguity and a long-tailed distribution. Advisors: Joseph Salmon, Maximilien Servajean, Alexis Joly.
• PhD in progress: Joaquim Estopinan, Species Conservation Status Prediction. Advisors: Alexis Joly, François Munoz, Maximilien Servajean, Pierre Bonnet.
• PhD in progress: Quentin Leroy, Interactive retrieval of non-common visual entities in large archives. Advisors: Olivier Buisson, Alexis Joly.
• PhD in progress: Lamia Djebour, Parallel Time Series Indexing and Retrieval with GPU architectures. Advisors: Reza Akbarinia, Florent Masseglia.

11.2.3 Juries

Members of the team participated to the following PhD or HDR committees:

• A. Joly: Etienne David, PhD, University of Avignon.
• A. Joly: Waleed RAGHEB, PhD, University of Montpellier.
• A. Joly: Alena Shilova, PhD, University of Bordeaux.
• F. Masseglia: Khaled Zaouk, PhD, Institut Polytechnique de Paris.
• F. Masseglia : Tao Peng, PhD, University of Aix-Marseille.
• E. Pacitti: Victorien Elvinger, PhD, Univ Lorraine, 2021.
• E. Pacitti: Lulian Sandu Popa, HDR, UVSQ - Université Paris-Saclay.
• P. Valduriez: Jorge Galicia Auyon, HDR, ENSMA, Univ Poitiers.
• A. Liutkus: Pritish Chandna (UPF Barcelona), Stylianos Mimilakis (Fraunhofer IDMT Ilmenau)

11.3.1 Internal or external Inria responsibilities

• Pl@ntNet Community management 2010-2021: A. Joly and H. Gresse spend several hours a week animating Pl@ntNet's user community. This includes: animating the community of developers using Pl@ntNet API (thousands of users, animating Pl@ntNet's social networks (twitter account, facebook account), managing the mailbox (contact@plantnet-project.org) and writing articles in the news section of the Pl@ntNet web site.
• F. Masseglia is co-head of the national project "1 Scientifique - 1 Classe, Chiche!"
• F. Masseglia is Member of the strategic committee of Fondation Blaise Pascal.

11.3.2 Articles and contents

• G. Heidsieck, E. Pacitti and F. Tardieu (INRAe) have co-produced a popularization video with Polytech' Montpellier on "The design of digital agriculture", May 2021.

11.3.3 Interventions

• A. Joly participated in a webinar organized by the European Citizen Science Association (ECSA) about the usage of Cos4Cloud services by citizen observatories.

International journals

• 15 articleA.Antoine Affouard, M.Mathias Chouet, J.-C.Jean-Christophe Lombardo, H.Hugo Gresse, H.Hervé Goëau, T.Titouan Lorieul, P.Pierre Bonnet and A.Alexis Joly. Customized e-floras: How to develop your own project on the Pl@ntNet platform.Biodiversity Information Science and Standards5September 2021, 1-4
  HALDOIback to text
• 16 articleM.-A.Marc-Antoine Benderra, A.Ainhoa Aparicio, J.Judith Leblanc, D.Demian Wassermann, E.Emmanuelle Kempf, G.Gilles Galula, M.Mélodie Bernaux, A.Anthony Canellas, T.Thomas Moreau, A.Ali Bellamine, J.-P.Jean-Philippe Spano, C.Christel Daniel, J.Julien Champ, F.Florence Canouï-Poitrine and J.Joseph Gligorov. Clinical Characteristics, Care Trajectories and Mortality Rate of SARS-CoV-2 Infected Cancer Patients: A Multicenter Cohort Study.Cancers1319September 2021, 4749
  HALDOI
• 17 articleE.Emmanuelle Blanc, P. M.Pierre M Barbillon, C.Christian Fournier, C.Christophe Lecarpentier, C.Christophe Pradal and J.Jérôme Enjalbert. Functional–Structural Plant Modeling Highlights How Diversity in Leaf Dimensions and Tillering Capability Could Promote the Efficiency of Wheat Cultivar Mixtures.Frontiers in Plant Science12September 2021, 734056
  HALDOI
• 18 articleH.Heraldo Borges, R.Reza Akbarinia and F.Florent Masseglia. Anomaly Detection in Time Series.Transactions on Large-Scale Data- and Knowledge-Centered Systems2021
  HALback to text
• 19 articleC.Christophe Botella, A.Alexis Joly, P.Pierre Bonnet, F.François Munoz and P.Pascal Monestiez. Jointly estimating spatial sampling effort and habitat suitability for multiple species from opportunistic presence‐only data.Methods in Ecology and Evolution125February 2021, 933-945
  HALDOI
• 20 articleB.Benjamin Deneu, M.Maximilien Servajean, P.Pierre Bonnet, C.Christophe Botella, F.François Munoz and A.Alexis Joly. Convolutional neural networks improve species distribution modelling by capturing the spatial structure of the environment.PLoS Computational Biology174April 2021, e1008856
  HALDOI
• 21 articleH.Hervé Goëau, P.Pierre Bonnet and A.Alexis Joly. AI-based Identification of Plant Photographs from Herbarium Specimens.Biodiversity Information Science and Standards5August 2021, 1-4
  HALDOIback to textback to text
• 22 articleG.Gaëtan Heidsieck, D.Daniel De Oliveira, E.Esther Pacitti, C.Christophe Pradal, F.Francois Tardieu and P.Patrick Valduriez. Cache-aware scheduling of scientific workflows in a multisite cloud.Future Generation Computer Systems1222021, 172-186
  HALDOI
• 23 articleP.Pavlos Kranas, B.Boyan Kolev, O.Oleksandra Levchenko, E.Esther Pacitti, P.Patrick Valduriez, R.Ricardo Jiménez-Peris and M.Marta Patiño-Martinez. Parallel Query Processing in a Polystore.Distributed and Parallel DatabasesFebruary 2021, 39
  HALDOI
• 24 articleL.Liliane Kunstmann, D.Débora Pina, F.Filipe Silva, A.Aline Paes, P.Patrick Valduriez, D.Daniel De Oliveira and M.Marta Mattoso. Online Deep Learning Hyperparameter Tuning based on Provenance Analysis.Journal of Information and Data Management125November 2021, 396-414
  HALDOI
• 25 articleO.Oleksandra Levchenko, B.Boyan Kolev, D.-E. E.Djamel-Edine Edine Yagoubi, R.Reza Akbarinia, F.Florent Masseglia, T.Themis Palpanas, P.Patrick Valduriez and D.Dennis Shasha. BestNeighbor: Efficient Evaluation of kNN Queries on Large Time Series Databases.Knowledge and Information Systems (KAIS)6322021, 349-378
  HALDOIback to text
• 26 articleN.Natalie Love, P.Pierre Bonnet, H.Hervé Goëau, A.Alexis Joly and S.Susan Mazer. Machine Learning Undercounts Reproductive Organs on Herbarium Specimens but Accurately Derives Their Quantitative Phenological Status: A Case Study of Streptanthus tortuosus.Plants1011November 2021, 2471
  HALDOIback to textback to text
• 27 articleC. A.Cyrille Ahmed Midingoyi, C.Christophe Pradal, A.Andreas Enders, D.Davide Fumagalli, H.Helene Raynal, M.Marcello Donatelli, I. N.Ioannis N. Athanasiadis, C.Cheryl Porter, G.Gerrit Hoogenboomi, D.Dean Holzworth, F.Frédérick Garcia, P.Peter Thorburn and P. M.Pierre Martres Martre. Crop2ML: An open-source multi-language modeling framework for the exchange and reuse of crop model components.Environmental Modelling and Software142August 2021
  HALDOI
• 28 articleB.Benjamin Pitchers, F.Frédéric Do, C.Christophe Pradal, L.Lydie Dufour and P.-É.Pierre-Éric Lauri. Apple tree adaptation to shade in agroforestry: an architectural approach.American Journal of Botany1085May 2021, 732-743
  HALDOI
• 29 articleN. C.Nigel C. A. Pitman, T.Tomomi Suwa, C.Carmen Ulloa Ulloa, J.James Miller, J.James Solomon, J.Juliana Philipp, C.Corine Vriesendorp, A.Abigail Derby Lewis, S.Sinem Perk, P.Pierre Bonnet, A.Alexis Joly, M.Mathias Tobler, J. H.Jason H. Best, J. P.John P. Janovec, K. C.Kevin C. Nixon, B. M.Barbara M. Thiers, M.Melissa Tulig, E. E.Edward E. Gilbert, R.Rafaela Campostrini Forzza, G.Geraldo Zimbrão, F. L.Fabiana Luiza Ranzato Filardi, R.Robert Turner, F.Fernando Zuloaga, M.Manuel Belgrano, C.Christian Zanotti, J. M.Jurriaan M. de Vos, E.Eduardo Hettwer Giehl, T. C.Timothy C. E. Paine, R.Rubens Texeira de Queiroz, K.Katya Romoleroux and E.Everton Hilo de Souza. Identifying gaps in the photographic record of the vascular plant flora of the Americas.Nature Plants72021, 1010-1014
  HALDOI
• 30 articleM.Maxime Ryckewaert, M.Maxime Metz, D.Daphné Héran, P.Pierre George, B.Bruno Grèzes-Besset, R.Reza Akbarinia, J. M.Jean Michel Roger and R.Ryad Bendoula. Massive spectral data analysis for plant breeding using parSketch-PLSDA method: Discrimination of sunflower genotypes.Biosystems Engineering210October 2021, 69-77
  HALDOIback to text
• 31 articleR.Rebecca Salles, E.Esther Pacitti, E.Eduardo Bezerra, F.Fabio Porto and E.Eduardo Ogasawara. TSPred: A framework for nonstationary time series prediction.Neurocomputing467January 2022, 197-202
  HALDOI
• 32 articleD. J.Daniel Junior Silva, E.Esther Pacitti, A.Aline Paes and D.Daniel De Oliveira. Provenance-and machine learning-based recommendation of parameter values in scientific workflows.PeerJ Computer Science7July 2021, e606
  HALDOIback to text
• 33 articleR.Renan Souza, L. G.Leonardo G Azevedo, V.Vítor Lourenço, E.Elton Soares, R.Raphael Thiago, R.Rafael Brandão, D.Daniel Civitarese, E. V.Emilio Vital Brazil, M.Marcio Moreno, P.Patrick Valduriez, M.Marta Mattoso, R.Renato Cerqueira and M. A.Marco A S Netto. Workflow Provenance in the Lifecycle of Scientific Machine Learning.Concurrency and Computation: Practice and ExperienceAugust 2021
  HALDOIback to text
• 34 articleR.Renan Souza, V.Vitor Silva, A.Alexandre Lima, D.Daniel De Oliveira, P.Patrick Valduriez and M.Marta Mattoso. Distributed in-memory data management for workflow executions.PeerJ Computer Science72021, e527
  HALDOI
• 35 articleH.Hirokazu Takahashi and C.Christophe Pradal. Root phenotyping: important and minimum information required for root modeling in crop plants.Breeding ScienceFebruary 2021
  HALDOI
• 36 articleP.Patrick Valduriez, R.Ricardo Jiménez-Peris and M. T.M. Tamer Özsu. Distributed Database Systems: The Case for NewSQL.Transactions on Large-Scale Data- and Knowledge-Centered Systems12670May 2021, 1-15
  HALDOI

International peer-reviewed conferences

• 37 inproceedingsF.Florent Abdelghafour, F.Florian Rançon, S.Shiyu Liu, J.Julien Champ, V.Vincent De Rudnicki, C.Christophe Guizard, H.Hervé Goëau, C.Claude Doussan, A.Alexis Joly, P.Pierre Bonnet and G.Gilles Rabatel. WeedElec: a robotic research platform for individual weed detection and selective electrical weeding.EPCA 2021 - 13th European Conference on Precision AgriculturePrecision agriculture ’21Budapest, HungaryWageningen Academic Publishers2021, 695-702
  HALDOIback to text
• 38 inproceedingsA.Antonio Castro, H.Heraldo Borges, R.Ricardo Campisano, E.Esther Pacitti, F.Fabio Porto, R.Rafaelli Coutinho and E.Eduardo Ogasawara. Generalização de Mineração de Sequências Restritas no Espaço e no Tempo.SBBD: Simpósio Brasileiro de Banco de DadosOnline, Brazil2021, 313-318
  HAL
• 39 inproceedingsB.Benjamin Deneu, A.Alexis Joly, P.Pierre Bonnet, M.Maximilien Servajean and F.François Munoz. How Do Deep Convolutional SDM Trained on Satellite Images Unravel Vegetation Ecology ?Lecture Notes in Computer ScienceICPR 2020 - 25th International Conference on Pattern Recognition12666Pattern Recognition. ICPR International Workshops and Challenges Virtual Event, January 10–15, 2021, Proceedings, Part VIMilan / Virtual, ItalySpringer2021, 148-158
  HALDOIback to text
• 40 inproceedingsL.Lamia Djebour, R.Reza Akbarinia and F.Florent Masseglia. ASAX : Segmentation adaptative basée sur la quantité d'information pour SAX.BDA 2021 - 37e Conférence sur la Gestion de Données - Principes, Technologies et ApplicationsParis, FranceOctober 2021
  HAL
• 41 inproceedingsC.Camille Garcin, A.Alexis Joly, P.Pierre Bonnet, J.-C.Jean-Christophe Lombardo, A.Antoine Affouard, M.Mathias Chouet, M.Maximilien Servajean, T.Titouan Lorieul and J.Joseph Salmon. Pl@ntNet-300K: a plant image dataset with high label ambiguity and a long-tailed distribution.NeurIPS 2021 - 35th Conference on Neural Information Processing SystemsVirtual Conference, FranceDecember 2021
  HALback to text
• 42 inproceedingsS. H.Sue Han Lee, H.Hervé Goëau, P.Pierre Bonnet and A.Alexis Joly. Conditional Multi-Task learning for Plant Disease Identification.ICPR 2020 - 25th International Conference on Pattern RecognitionProceedings of ICPR 2020, 25th International Conference on Pattern RecognitionMilan, ItalyIEEE2021, 3320-3327
  HALDOIback to text
• 43 inproceedingsA.Antoine Liutkus, O.Ondřej Cífka, S.-L.Shih-Lun Wu, U.Umut Şimşekli, Y.-H.Yi-Hsuan Yang and G.Gael Richard. Relative Positional Encoding for Transformers with Linear Complexity.ICML 2021 - 38th International Conference on Machine LearningProceedings of the 38th International Conference on Machine LearningVirtual Only, United StatesJuly 2021
  HAL
• 44 inproceedingsT.Titouan Lorieul, E.Elijah Cole, B.Benjamin Deneu, M.Maximilien Servajean, P.Pierre Bonnet and A.Alexis Joly. Overview of GeoLifeCLEF 2021: Predicting species distribution from 2 million remote sensing images.CEUR workshop ProceedingsWorking Notes of CLEF 2021 - Conference and Labs of the Evaluation Forum2936Working Notes of CLEF 2021 - Conference and Labs of the Evaluation ForumBucarest, Romania2021, 1451-1462
  HAL
• 45 inproceedingsD.Débora Pina, L.Liliane Kunstmann, D.Daniel De Oliveira, P.Patrick Valduriez and M.Marta Mattoso. Provenance Supporting Hyperparameter Analysis in Deep Neural Networks.IPAW 2020-2021 - 8th and 9th International Provenance and Annotation WorkshopLecture Notes in Computer ScienceProvenance and Annotation of Data and Processes12839London, United KingdomSpringerJuly 2021, 20-38
  HALback to text
• 46 inproceedingsD.Daniel Rosendo, A.Alexandru Costan, G.Gabriel Antoniu, M.Matthieu Simonin, J.-C.Jean-Christophe Lombardo, A.Alexis Joly and P.Patrick Valduriez. Reproducible Performance Optimization of Complex Applications on the Edge-to-Cloud Continuum.Cluster 2021 - IEEE International Conference on Cluster ComputingPortland, OR, United StatesSeptember 2021
  HALback to text
• 47 inproceedingsR. A.Rodrigo A. P. Silva, E.Esther Pacitti, Y. Y.Yuri Y. Frota and D.Daniel De Oliveira. Análise de Desempenho da Distribuição de Workflows Científicos em Nuvens com Restrições de Confidencialidade.Workshop on Computer and Communication Systems Performance (WPerformance 2021)Online, BrazilJuly 2021, 12
  HALDOI
• 48 inproceedingsR. M.Rômulo M. Silva, D.Débora Pina, L.Liliane Kunstmann, D.Daniel De Oliveira, P.Patrick Valduriez, A. L.Alvaro L. G. A. Coutinho and M.Marta Mattoso. Capturing Provenance to Improve the Model Training of PINNs: first handon experiences with Grid5000.42nd Ibero-Latin-American Congress on Computational Methods in Engineering and 3rd Pan American Congress on Computational MechanicsRio de Janeiro, BrazilNovember 2021, 1-7
  HAL
• 49 inproceedingsC.Chao Zhang, R.Reza Akbarinia and F.Farouk Toumani. Efficient Incremental Computation of Aggregations over Sliding Windows.KDD 2021 - 27th ACM SIGKDD Conference on Knowledge Discovery and Data MiningSingapore (Virtual), Singapore2021, 2136-2144
  HALDOI

National peer-reviewed Conferences

• 50 inproceedingsD.Daniel Rosendo, A.Alexandru Costan, G.Gabriel Antoniu and P.Patrick Valduriez. Enabling Reproducible Analysis of Complex Workflows on the Edge-to-Cloud Continuum.BDA 2021 - 37e Conférence sur la Gestion de Données - Principes, Technologies et ApplicationsProceedings of the BDA 2021 conferenceParis, FranceOctober 2021
  HALback to text
• 51 inproceedingsC.Chao Zhang, R.Reza Akbarinia and F.Farouk Toumani. Efficient Incremental Computation of Aggregations over Sliding Windows.BDA 2021 - 37e Conférence sur la Gestion de Données - Principes, Technologies et ApplicationsVirtual, FranceColegio de Médicos y Cirujanos de Costa RicaOctober 2021
  HALDOIback to text

Conferences without proceedings

• 52 inproceedingsA.Alexis Joly, A.Antoine Affouard, M.Mathias Chouet, B.Benjamin Deneu, J.Joaquim Estopinan, H.Hervé Goëau, H.Hugo Gresse, J.-C.Jean-Christophe Lombardo, T.Titouan Lorieul, F.François Munoz, M.Maximilien Servajean and P.Pierre Bonnet. Pl@ntNet, ten years of automatic plant identification and monitoring.IUCN - Congrès mondial de la natureMarseille, FranceSeptember 2021
  HAL

Scientific book chapters

• 53 inbookA.Alexis Joly, H.Hervé Goëau, S.Stefan Kahl, L.Lukáš Picek, T.Titouan Lorieul, E.Elijah Cole, B.Benjamin Deneu, M.Maximilien Servajean, A.Andrew Durso, I.Isabelle Bolon, H.Hervé Glotin, R.Robert Planqué, R. R.Rafael Ruiz de Castañeda, W.-P.Willem-Pier Vellinga, H.Holger Klinck, T.Tom Denton, I.Ivan Eggel, P.Pierre Bonnet and H.Henning Müller. Overview of LifeCLEF 2021: an evaluation of Machine-Learning based Species Identification and Species Distribution Prediction.12880Experimental IR Meets Multilinguality, Multimodality, and InteractionLecture Notes in Computer ScienceSpringer International PublishingSeptember 2021, 371-393
  HALDOIback to text

Doctoral dissertations and habilitation theses

• 54 thesisA.Antoine Liutkus. Principled methods for mixtures processing.Université de MontpellierFebruary 2022
  HAL

Reports & preprints

• 55 miscE.Evgenii Chzhen, C.Christophe Denis, M.Mohamed Hebiri and T.Titouan Lorieul. Set-valued classification -- overview via a unified framework.March 2021
  HAL
• 56 miscC.Camille Garcin, M.Maximilien Servajean, A.Alexis Joly and J.Joseph Salmon. Stochastic smoothing of the top-K calibrated hinge loss for deep imbalanced classification.February 2022
  HAL

Other scientific publications

• 57 miscG.Gabriel Antoniu, P.Patrick Valduriez, H.-C.Hans-Christian Hoppe and J.Jens Krüger. Towards Integrated Hardware/Software Ecosystems for the Edge-Cloud-HPC Continuum.2021
  HALDOIback to text
• 58 inproceedingsD.Daniel Rosendo, A.Alexandru Costan, G.Gabriel Antoniu and P.Patrick Valduriez. E2Clab: Reproducible Analysis of Complex Workflows on the Edge-to-Cloud Continuum.IPDPS 2021 - 35th IEEE International Parallel and Distributed Processing SymposiumVirtual, FranceMay 2021
  HAL

Scientific popularization

• 59 articleP.Patrick Valduriez. Making the Right Move to Senior Researcher: some challenges and hints.SIGMOD record502June 2021, 30-32
  HALDOI