2025Activity​‌ reportProject-TeamIROKO

Unified data–model–workflow​‌ services.

Environmental science pipelines​​ increasingly combine heterogeneous data​​​‌ (e.g., images, omics, epidemiology,​ climate) with heterogeneous models​‌ (statistical, machine learning, mechanistic)​​ and complex workflows. Yet​​​‌ current solutions remain largely​ domain-specific and ad hoc,​‌ making it difficult to​​ connect artifacts, reproduce results,​​​‌ and reuse components across​ projects. Our goal is​‌ to provide integrated data​​ and model management together​​​‌ with workflow services that​ can interoperate with established​‌ environments such as Galaxy​​ 71 and OpenAlea 95​​, as well as​​​‌ distributed execution engines. Building‌ on our LifeSWS initiative‌​‌ 72, we aim​​ to treat all scientific​​​‌ artifacts (datasets, models, metadata,‌ workflow components, intermediate results)‌​‌ as first-class citizens, searchable​​ via catalogs and executable​​​‌ through standardized interfaces. Compared‌ to existing platforms (the‌​‌ closest being CyVerse 83​​), our focus is​​​‌ on tighter integration of‌ model life-cycle management, provenance,‌​‌ and caching to support​​ end-to-end scientific investigations.

Scalable​​​‌ time-series analytics at climate‌ scale.

Many environmental questions‌​‌ rely on large collections​​ of time series and​​​‌ call for motif discovery,‌ clustering, and anomaly detection.‌​‌ At scale, naive distributed​​ adaptations can be inefficient​​​‌ due to communication and‌ synchronization costs 100.‌​‌ We will therefore design​​ distribution-aware algorithms for time-series​​​‌ analytics, with a particular‌ focus on anomaly detection‌​‌ in large climate datasets​​ using Matrix Profile ideas​​​‌ 101, and on‌ implementations that can leverage‌​‌ modern distributed infrastructures (e.g.,​​ Spark) without sacrificing usability​​​‌ for domain partners.

• Mid-term:‌ a first operational version‌​‌ of LifeSWS enabling integrated​​ artifact search and workflow​​​‌ execution across heterogeneous environments;‌ a distributed Matrix-Profile-based anomaly‌​‌ detection prototype validated on​​ large environmental time series.​​​‌
• Long-term: a production-grade, scalable‌ service stack (catalog, provenance,‌​‌ caching, scheduling) enabling reproducible​​ analyses across data modalities​​​‌ and models; a toolbox‌ of distributed time-series operators‌​‌ usable in multiple environmental​​ and One Health contexts.​​​‌

Cooperative learning in citizen​​ science.

Platforms such as​​​‌ Pl@ntNet and iNaturalist continuously‌ improve their identification models‌​‌ from community-produced observations and​​ revisions. This cooperative learning​​​‌ loop is powerful but‌ raises new issues: sparse‌​‌ and opportunistic revisions, strong​​ imbalance across taxa/regions/users, and​​​‌ extreme scale (tens of‌ millions of observations; tens‌​‌ of thousands of classes).​​ Standard crowdsourcing inference tools​​​‌ (e.g., Bayesian aggregation) are‌ not directly applicable at‌​‌ this scale 86.​​ We will develop end-to-end​​​‌ human-in-the-loop models that represent‌ user behavior and its‌​‌ impact on training dynamics,​​ building on our initial​​​‌ contributions 88 and leveraging‌ modern approaches to detect‌​‌ label issues 94.​​ A key aim is​​​‌ to prevent negative feedback‌ loops and to learn‌​‌ principled user weighting strategies​​ that remain robust under​​​‌ sparsity and imbalance.

Bias-aware‌ species distribution models from‌​‌ opportunistic data.

To monitor​​ biodiversity under rapid global​​​‌ change 85, species‌ distribution models (SDMs) increasingly‌​‌ rely on citizen science​​ data, whose scale is​​​‌ unmatched but whose biases‌ are substantial 73,‌​‌ 79. We will​​ continue to develop statistically​​​‌ grounded bias-correction methods 78‌, 77 and extend‌​‌ them to modern AI-based​​ SDMs 76 and to​​​‌ Bayesian dynamic SDMs 75‌. We will also‌​‌ address the fundamental limitation​​ of presence-only data by​​​‌ inferring absences from visit‌ histories and multi-species information‌​‌ 97, and by​​ modeling observer profiles (persistent​​​‌ users, learners, taxonomic preferences)‌ to better disentangle detectability‌​‌ from ecological signals.

Uncertainty,​​ trust, and interpretability.

Reliable​​​‌ downstream decisions require explicit‌ management of predictive uncertainty.‌​‌ We will build on​​ our work on set-valued​​​‌ classification and abstention mechanisms‌ 91, 82 and‌​‌ investigate uncertainty quantification and​​​‌ propagation for structured biodiversity​ predictions (assemblages, indicators, abundance​‌ maps). This includes generic​​ tools such as conformal​​​‌ prediction 96, 81​ and scalable Bayesian approximations​‌ 70. Finally, we​​ will strengthen user trust​​​‌ through transparency and interpretability​ mechanisms, extending prior work​‌ on user-facing uncertainty and​​ interactive features in Pl@ntNet​​​‌ 90.

• Mid-term: new​ scalable HITL (Human In​‌ The Loop) models evaluated​​ on real Pl@ntNet-style revision​​​‌ streams; open-source SDM training​ components with improved bias​‌ handling; first results on​​ uncertainty propagation for biodiversity​​​‌ indicators.
• Long-term: integration of​ HITL and uncertainty-aware decision​‌ modules into Pl@ntNet/GeoPl@ntNet-like services;​​ robust bias-aware dynamic SDM​​​‌ pipelines for long-term monitoring​ and scenario analysis.

Multimodal foundation models​​​‌ for biodiversity and agro-ecology​ monitoring.

High-resolution monitoring of​‌ biotic components remains challenging​​ despite major advances in​​​‌ sensors and AI 74​. We will study​‌ learning strategies that combine​​ smartphone observations, scientific imaging​​​‌ workflows, drones, remote sensing,​ and environmental covariates (e.g.,​‌ bioclimatic layers 80)​​ into multimodal pipelines. Because​​​‌ fully end-to-end multimodal training​ can be costly, we​‌ will emphasize self-supervised and​​ foundation-model approaches, then reuse​​​‌ learned representations for downstream​ ecological tasks. We will​‌ also prioritize interpretability, combining​​ transparency principles with post-hoc​​​‌ explanations (e.g., Shapley-based methods​ 98).

Multivariate time​‌ series and scalable similarity.​​

Environmental and One Health​​​‌ applications increasingly involve multivariate​ time series, where variables​‌ interact across time and​​ scales. We will develop​​​‌ parallel anomaly detection and​ similarity search methods (including​‌ kNN Matrix Profile variants)​​ building on our prior​​​‌ distributed indexing and analytics​ experience 100, 92​‌. We will also​​ investigate visualization and representation​​​‌ learning for complex multivariate​ temporal data to support​‌ interactive exploration by domain​​ experts.

Biodiversity trajectories and​​​‌ community structure.

Predicting biodiversity​ trajectories requires models that​‌ capture long-term dependencies and​​ integrate heterogeneous historical evidence.​​​‌ We will investigate deep​ architectures (including transformer-based models)​‌ for integrative forecasting 89​​, and contrast them​​​‌ with interpretable dynamic models​ grounded in ecological mechanisms​‌ (e.g., Bayesian DSDMs 75​​). At the community​​​‌ level, we will leverage​ graph-based approaches to analyze​‌ and predict species assemblages​​ via co-occurrence networks, relating​​​‌ network structure to dispersal,​ filtering, and interactions 93​‌, and validating on​​ vegetation survey datasets 99​​​‌.

• Mid-term: prototypes for​ multi-species monitoring from complex​‌ imagery and environmental covariates;​​ multivariate time-series analytics components;​​​‌ first case studies on​ forecasting short-term biodiversity trends.​‌
• Long-term: reusable multimodal foundation​​ models shared openly; operational​​​‌ toolchains for multiscale forecasting​ and scenario analysis; new​‌ graph-based methods to predict​​ and interpret species assemblages.​​​‌

7.2.1 GeoPl@ntNet

Participants: Lukas​​​‌ Picek, César Leblanc‌, Benjamin Deneu,‌​‌ Rémi Palard, Thomas​​ Paillot, Christophe Botella​​​‌, Alexis Joly.‌

GeoPl@ntNet is a new,‌​‌ large-scale web application developed​​ in the context of​​​‌ the Pl@ntNet platform for‌ the exploration, analysis, and‌​‌ dissemination of plant biodiversity​​ information, offering an unprecedented​​​‌ combination of taxonomic coverage,‌ spatial extent, and spatial‌​‌ resolution. The application provides​​ high-resolution distribution maps (50​​​‌ $\times$ 50 m) for‌ more than 15,000 plant‌​‌ species across the entire​​ European continent, making it​​​‌ one of the most‌ comprehensive operational systems currently‌​‌ available for plant biodiversity​​ mapping at this scale.​​​‌ GeoPl@ntNet relies on state-of-the-art‌ deep learning–based species distribution‌​‌ models that integrate heterogeneous​​​‌ environmental data—such as satellite​ imagery, climatic variables, land-use​‌ information, and topography—with millions​​ of in situ plant​​​‌ observations collected through the​ Pl@ntNet platform. Beyond interactive​‌ visualization, the application allows​​ users to explore regions​​​‌ of interest, compute biodiversity​ indicators (including protected, invasive,​‌ and endemic species), and​​ access detailed, spatially explicit​​​‌ reports to support research,​ conservation planning, and territorial​‌ management. A key feature​​ of GeoPl@ntNet is the​​​‌ open availability of its​ outputs: all species distribution​‌ maps are made freely​​ downloadable, fostering transparency, reuse,​​​‌ and integration into external​ scientific studies, public policies,​‌ and operational workflows. By​​ combining continental-scale coverage, fine​​​‌ spatial resolution, and open​ data dissemination within a​‌ single platform, GeoPl@ntNet represents​​ a unique operational contribution​​​‌ to large-scale plant biodiversity​ monitoring and decision support.​‌ The application is already​​ used by more than​​​‌ 1K users per month.​

7.2.2 PROMISE

Participants: Reza​‌ Akbarinia, Benoit Lange​​, Florent Masseglia.​​​‌

The objective of the​ PROMISE (PROfessional coMmunIty network​‌ on antimicrobial reSistancE) project​​ 87 is to build​​​‌ a large data warehouse​ for managing and analyzing​‌ antimicrobial resistance (AMR) data.​​ The PROMISE platform, of​​​‌ the same name, is​ a multi-cloud data management​‌ and analytics platform developed​​ in the context of​​​‌ the PROMISE project to​ support One Health surveillance​‌ and research on antimicrobial​​ resistance. The platform integrates​​​‌ data from the human,​ animal and environmental sectors.​‌ At present, the data​​ handled by the platform​​​‌ are aggregated (no personal​ data) and largely derived​‌ from public sources. PROMISE​​ relies on a modular​​​‌ architecture organized into five​ independent "bubbles" (diffusion, query,​‌ storage, administration and processing)​​ that can be deployed​​​‌ on any cloud. Services​ are containerized (Docker) and​‌ orchestrated with Kubernetes; inter-service​​ communication is performed through​​​‌ REST APIs, and WebSockets​ are used for notifications.​‌

The diffusion bubble provides​​ both the web user​​​‌ interface and the API​ entry point, with a​‌ React-based viewer and a​​ Quarkus (Java) gateway that​​​‌ routes requests to the​ relevant services. The administration​‌ bubble manages authentication and​​ observability (monitoring and metrics​​​‌ collection). The query bubble​ normalizes user requests and​‌ aggregates results, while the​​ storage bubble isolates raw​​​‌ data on the providers’​ infrastructures, translates normalized queries​‌ into database-specific queries, and​​ returns aggregated outputs. Current​​​‌ storage connectors support PostgreSQL,​ InfluxDB and MongoDB. The​‌ processing bubble orchestrates analytics​​ over aggregated time series​​​‌ and supports correlation modules​ implemented in Python and​‌ connected to an event​​ bus. Finally, an HDS-oriented​​​‌ deployment option is being​ investigated to enable the​‌ use of more sensitive​​ health data while preserving​​​‌ a strict separation between​ raw data and aggregated​‌ outputs.

Contact: Reza Akbarinia​​

8.1.1 A​​ Logic-Based Approach for Knowledge​​​‌ Graph Data Integration

Participants:‌ Fabio Porto, Patrick‌​‌ Valduriez.

In the​​ context of the Dinizia​​​‌ Inria associated team with‌ Brazil, we started a‌​‌ collaboration with the Boreal​​ Inria team to study​​​‌ the combination of a‌ knowledge graph with rule-based‌​‌ reasoning. In particular, we​​ are interested in leveraging​​​‌ the InteGraal framework developed‌ within the Boreal team,‌​‌ which enables semantic integration​​ and reasoning over heterogeneous​​​‌ data sources. In this‌ context, we proposed Gypscie-KG‌​‌ 55, an ML​​ (Machine Learning) system that​​​‌ combines data integration, rule-based‌ reasoning, and prediction services‌​‌ to provide semantic access​​ to domain knowledge using​​​‌ a knowledge graph. In‌ addition to providing integration‌​‌ of heterogeneous ML data​​ within a knowledge graph,​​​‌ we explore the use‌ of logic-based declarative techniques‌​‌ to enable reasoning and​​ semantic querying over ML​​​‌ data.

8.1.2 Federated Learning‌

Participants: Patrick Valduriez.‌​‌

Federated Learning (FL) is​​ a promising distributed machine​​​‌ learning approach that enables‌ collaborative training of a‌​‌ global model using multiple​​ edge devices. The data​​​‌ distributed among the edge‌ devices is highly heterogeneous.‌​‌ Thus, FL faces the​​ challenge of data distribution​​​‌ and heterogeneity, where non-independent‌ and identically distributed (non-IID)‌​‌ data across edge devices​​ may result in a​​​‌ significant accuracy drop. Furthermore,‌ the limited computation and‌​‌ communication capabilities of edge​​ devices increase the likelihood​​​‌ of stragglers, thus leading‌ to slow model convergence.‌​‌ To address this problem,​​ we proposed the FedDHAD​​​‌ FL framework 26,‌ which comes with two‌​‌ novel methods: Dynamic Heterogeneous​​ model aggregation (FedDH) and​​​‌ Adaptive Dropout (FedAD). The‌ combination of these two‌​‌ methods makes FedDHAD significantly​​ outperform state-of-the-art solutions in​​​‌ terms of accuracy (up‌ to 6.7% higher), efficiency‌​‌ (up to 2.02 times​​ faster), and computation cost​​​‌ (up to 15.0% smaller).‌

8.1.3 Distributed Web Infrastructure‌​‌ for Integrated Pest Management​​

Participants: Christophe Pradal.​​​‌

Crop protection and pest‌ management are major economic‌​‌ and environmental concerns throughout​​ Europe. The consultation of​​​‌ decision support systems (DSS)‌ to guide decisions relating‌​‌ to Integrated Pest Management​​ (IPM) is one of​​​‌ the key principles of‌ IPM, reducing the ambiguity‌​‌ around potential risks to​​ crop health. Pests in​​​‌ this context include invertebrate‌ pests, weeds and pathogens.‌​‌

In 63, to​​​‌ facilitate the use of​ these models, two Application​‌ Programming Interfaces (APIs) were​​ designed to access catalog​​​‌ of DSS models and​ European online weather data​‌ sources. While these APIs​​ are integrated into the​​​‌ IPM Decisions Platform (​IPM Decisions Platform),​‌ they are also open​​ source, allowing other crop​​​‌ protection and farm management​ software to inspect, download,​‌ modify, install, run, and​​ use them.

The scientific​​​‌ platform OpenAlea provides a​ new service, the IPM​‌ Decision Factory, that enables​​ DSS researchers and developers​​​‌ to advance, combine and​ create DSS interactively into​‌ its scientific workflow management​​ system. These workflows are​​​‌ then automatically transformed into​ web services to be​‌ readily integrated into the​​ IPM Decisions platform. This​​​‌ ensures that new DSS​ have access to required​‌ weather data and can​​ be made readily accessible​​​‌ across Europe, for validation​ and use. OpenAlea.EpyMix 25​‌ is a model describing​​ canopy growth and epidemic​​​‌ dynamics on species mixture​ that has been integrated​‌ into the IPM Decision​​ platform to understand how​​​‌ weather data, provided by​ the platform, and wheat-based​‌ crop mixtures are a​​ promising strategy to improve​​​‌ disease management.

8.2.1 Event Detection​‌ in Time Series

Participants:​​ Esther Pacitti, Fabio​​​‌ Porto, Rebecca Salles​.

Event detection in​‌ time series is a​​ basic function in surveillance​​​‌ and monitoring systems and​ has been extensively explored​‌ over the years.

The​​ new book 56 published​​​‌ by Springer and authored​ by Eduardo Ogasawara (CEFET-RJ,​‌ Brazil), Rebecca Salles (Iroko),​​ Fabio Porto (LNCC, Brazil)​​​‌ and Esther Pacitti (Iroko),​ reflects our productive collaboration​‌ with Brazil in the​​ context of the Dinizia​​​‌ associated team. It provides​ a general taxonomy for​‌ event detection according to​​ the specific event types:​​​‌ anomaly detection, change-point, and​ motif discovery. It discusses​‌ state-of-the-art metric evaluations for​​ event detection methods and​​​‌ on online event detection,​ including the challenges of​‌ incremental and adaptive learning.​​

Anomaly detection methods implicitly​​​‌ define detection criteria, such​ as deviation measures, filter​‌ thresholds, and candidate anomaly​​ selection strategies. Choosing inappropriate​​​‌ criteria results in inaccurate​ outputs, generating spurious alerts​‌ or missing events. Adjusting​​ these criteria is essential​​​‌ for monitoring systems. To​ address this challenge, we​‌ explored the fine-tuning of​​ deviation measures, filter thresholds,​​​‌ and candidate selection strategies​ 52. Experimental results​‌ show that the proper​​ choice of criteria significantly​​​‌ improves anomaly detection performance,​ often with greater impact​‌ than changing the detection​​ methods.

Concept drift detection​​​‌ (CDD) is the general​ problem of identifying significant​‌ changes in streaming data​​ distribution over time. Current​​​‌ CDD methods face challenges​ in large-scale, multivariate datasets,​‌ where single drift detectors​​ (DD) often fail to​​​‌ capture variable interdependencies. While​ ensemble drift detectors (EDD)​‌ are usually adopted to​​ mitigate the adoption of​​​‌ a single DD, EDD​ may suffer when detections​‌ do not converge. This​​ misalignment can cause voting​​​‌ mechanisms to neglect critical​ intervals with high detection​‌ rates. To address this​​ issue, we proposed a​​​‌ fuzzy ensemble drift detector​ (FEDD) 44 that integrates​‌ unsupervised threshold voting with​​ fuzzy logic to provide​​ time tolerance and reconcile​​​‌ minor temporal misalignments in‌ drift detection. Our evaluation‌​‌ shows that FEDD outperforms​​ existing approaches by improving​​​‌ detection robustness and coverage.‌

8.2.2 Scalable Multivariate Anomaly‌​‌ Detection

Participants: Reza Akbarinia​​, Benoit Lange,​​​‌ Florent Masseglia, Esther‌ Pacitti, Rebecca Salles‌​‌.

The continuous monitoring​​ of dynamic processes generates​​​‌ vast amounts of streaming‌ multivariate time series data.‌​‌ Detecting anomalies within them​​ is crucial for real-time​​​‌ identification of significant events,‌ such as environmental phenomena,‌​‌ security breaches, or system​​ failures, which can critically​​​‌ impact sensitive applications. Despite‌ significant advances in univariate‌​‌ time series anomaly detection,​​ scalable and efficient solutions​​​‌ for online detection in‌ multivariate streams remain underexplored.‌​‌ This challenge becomes increasingly​​ prominent with the growing​​​‌ volume and complexity of‌ multivariate time series data‌​‌ in streaming scenarios.

In​​ 33, we provide​​​‌ the first structured survey‌ primarily focused on scalable‌​‌ and online anomaly detection​​ techniques for multivariate time​​​‌ series, offering a comprehensive‌ taxonomy. Additionally, we introduce‌​‌ the Online Distributed Outlier​​ Detection (2OD) methodology, a​​​‌ novel well-defined and repeatable‌ process designed to benchmark‌​‌ the online and distributed​​ execution of anomaly detection​​​‌ methods. Experimental results with‌ both synthetic and real-world‌​‌ datasets, covering up to​​ hundreds of millions of​​​‌ observations, demonstrate that a‌ distributed approach can enable‌​‌ centralized algorithms to achieve​​ significant computational efficiency gains,​​​‌ averaging tens and reaching‌ up to hundreds in‌​‌ speedup, without compromising detection​​ accuracy.

8.2.3 Detecting Anomalies​​​‌ with Any Duration in‌ Climate Time Series

Participants:‌​‌ Reza Akbarinia, Guillaume​​ Coulaud, Florent Masseglia​​​‌.

Detecting abnormal climate‌ events across temporal and‌​‌ spatial scales is crucial​​ to the understanding of​​​‌ local and regional climate‌ trends. Existing methods often‌​‌ depend on prior knowledge​​ about the timing, location,​​​‌ or duration of such‌ events, limiting their versatility.‌​‌ In 15, we​​ propose ClimBurst, an approach​​​‌ to detect climate bursts‌ (unusually high or low‌​‌ values of climate variables)​​ without prior assumptions about​​​‌ their temporal duration. ClimBurst‌ offers the ability to:‌​‌ (a) identify climate bursts​​ of any duration within​​​‌ the time series of‌ single locations, (b) link‌​‌ climate bursts across neighboring​​ locations, and (c) analyze​​​‌ the spatio‐temporal propagation of‌ these anomalies. Applying ClimBurst‌​‌ to sea surface temperature​​ data from the Mediterranean​​​‌ Sea (1960–2021) shows some‌ detected hot bursts and‌​‌ anomalies coincide in time​​ with known severe marine​​​‌ heatwaves. ClimBurst also shows‌ how detected hot (cold)‌​‌ bursts are spatio‐temporally connected​​ and these connected bursts​​​‌ have increased (decreased) in‌ duration, intensity, spatial extent‌​‌ and frequency historically.

In​​ 40, we propose​​​‌ a demonstration of ClimBurst‌ allowing users to interact‌​‌ directly with our system​​ to see both a​​​‌ summary showing the presence/absence‌ of bursts over a‌​‌ user-specified year and spatial​​ range. The demonstration will​​​‌ also allow users to‌ perform time-travel queries to‌​‌ see how bursts propagate​​ over space and time.​​​‌

8.2.4 Energy Efficient Time‌ Series Anomaly Detection

Participants:‌​‌ Reza Akbarinia, Benoit​​ Lange, Florent Masseglia​​​‌, Esther Pacitti,‌ Rebecca Salles.

Traditionally,‌​‌ choosing an anomaly detection​​​‌ method for a given​ application is mainly driven​‌ by detection accuracy and​​ runtime. However, with the​​​‌ rapid evolution of hardware​ and connected devices, massive​‌ amounts of time series​​ data are produced, and​​​‌ the real-time analysis of​ such time series brings​‌ new demands not only​​ for accurate and scalable​​​‌ solutions, but also for​ energy consumption management. In​‌ this scenario, any improvement​​ in energy efficiency can​​​‌ have a considerable impact​ on both the environmental​‌ footprint and the monetary​​ expenses. In 53,​​​‌ we address the problem​ of benchmarking time series​‌ anomaly detection methods based​​ on the trade-off between​​​‌ accuracy, runtime, and energy​ consumption. We introduce a​‌ new metric for evaluating​​ relative energy efficiency performance,​​​‌ called saveUp, and provide​ a novel methodology, inspired​‌ by skyline queries, for​​ benchmarking methods based on​​​‌ a more comprehensive set​ of metrics, including peak​‌ power usage and total​​ energy consumption. Experimental results​​​‌ based on large datasets​ show that our methodology​‌ is useful for selecting​​ the methods that provide​​​‌ the best performance with​ the lowest energy impacts.​‌ Moreover, results indicate that​​ speedup and saveUp are​​​‌ not always directly correlated​ as believed a priori,​‌ and sometimes it is​​ best to "take it​​​‌ slow" in favor of​ green applications.

8.2.5 Extending​‌ Matrix Profile for Seasonal​​ Anomaly Detection

Participants: Reza​​​‌ Akbarinia, Guillaume Coulaud​, Florent Masseglia.​‌

Seasonal time series analysis​​ is fundamental in domains​​​‌ such as climate science,​ where detecting and understanding​‌ anomalies, patterns, and data​​ changes are essential. The​​​‌ classical Matrix Profile approach​ does not consider the​‌ data’s seasonality, failing to​​ detect seasonal anomalies and​​​‌ patterns. In 60,​ we propose the Interval​‌ Matrix Profile (IMP), a​​ novel extension of the​​​‌ Matrix Profile specifically designed​ for analyzing periodic and​‌ seasonal time series data.​​ The Interval Matrix Profile​​​‌ enables flexible interval-based comparisons​ across seasons, allowing the​‌ detection of anomalies that​​ conventional approaches miss. We​​​‌ further propose the constrained​ k Nearest Neighbor Interval​‌ Matrix Profile, designed to​​ identify anomalies that may​​​‌ appear across multiple periods,​ a common characteristic of​‌ abnormal climate events and​​ extreme weather phenomena. Our​​​‌ approach leverages a scalable​ block-based algorithm that achieves​‌ significant performance gains through​​ caching, vectorization, and parallelism.​​​‌ Additionally, we introduce a​ novel methodology to detect​‌ the first or last​​ occurrence of a pattern,​​​‌ enabling the discovery of​ pattern emergence or disappearance​‌ within seasonal time series.​​ The algorithms are demonstrated​​​‌ in case studies on​ temperature climate time series.​‌ They effectively capture seasonal​​ anomalies and find pattern​​​‌ disappearance. Our results illustrate​ that the IMP consistently​‌ outperforms the classical Matrix​​ Profile both in the​​​‌ accuracy of seasonal anomaly​ detection and in computational​‌ efficiency.

8.3.1 Learning Ecological Structure​ with Large Language Models​‌

Participants: César Leblanc,​​ Hervé Goëau, Maximilien​​​‌ Servajean, Alexis Joly​, Diego Marcos,​‌ Pierre Bonnet.

This​​ research axis explores how​​​‌ large language models (LLMs)​ can be adapted to​‌ capture and exploit structured​​ ecological knowledge, with a​​ focus on plant communities​​​‌ and functional traits. By‌ transferring ideas from natural‌​‌ language processing to ecology,​​ these works investigate how​​​‌ latent structure in species‌ assemblages and unstructured textual‌​‌ resources can be leveraged​​ to improve biodiversity understanding​​​‌ and modeling.

In 23‌, the team introduces‌​‌ an approach inspired by​​ language modeling to learn​​​‌ the “syntax” of plant‌ assemblages, treating abundance-ordered species‌​‌ lists as ecological sequences.​​ Trained on more than​​​‌ 10,000 European plant species,‌ the model captures latent‌​‌ associations shaped by environmental​​ constraints, dispersal processes, and​​​‌ species interactions. The learned‌ representations can be fine-tuned‌​‌ for multiple downstream tasks,​​ including predicting missing species​​​‌ in assemblages and classifying‌ habitat types, where the‌​‌ method consistently outperforms co-occurrence-based​​ models, expert systems, and​​​‌ standard neural networks. This‌ work demonstrates how sequence-based‌​‌ modeling provides a powerful​​ and flexible framework for​​​‌ representing plant community structure.‌

Complementing this community-level perspective,‌​‌ 27 focuses on species-level​​ functional information and addresses​​​‌ the challenge of assembling‌ large trait databases. Leveraging‌​‌ the information extraction capabilities​​ of large language models,​​​‌ this work proposes a‌ fully automatic pipeline to‌​‌ extract plant morphological traits​​ from unstructured online textual​​​‌ descriptions. The approach successfully‌ reconstructs expert-curated species–trait matrices‌​‌ with high accuracy, showing​​ that LLMs can transform​​​‌ heterogeneous textual resources into‌ structured ecological knowledge at‌​‌ scale, albeit with current​​ limitations linked to data​​​‌ availability.

Together, these contributions‌ illustrate the potential of‌​‌ large language models to​​ bridge different levels of​​​‌ ecological organization, from individual‌ traits to species assemblages,‌​‌ and to open new​​ avenues for scalable, data-driven​​​‌ biodiversity modeling, mapping, and‌ conservation science.

8.3.2 Scalable‌​‌ Plant Vision Models for​​ Operational Monitoring

Participants: Hervé​​​‌ Goëau, Vincent Espitalier‌, Alexis Joly,‌​‌ Pierre Bonnet.

This​​ research axis investigates how​​​‌ large-scale plant vision models‌ can be designed and‌​‌ adapted for operational monitoring​​ tasks, with a strong​​​‌ emphasis on scalability, robustness,‌ and reduced annotation requirements‌​‌ in real-world conditions.

In​​ 20, the team​​​‌ addresses the early detection‌ of invasive alien plant‌​‌ species along roadsides, a​​ major vector for biological​​​‌ invasions. Rather than relying‌ on object detection or‌​‌ segmentation pipelines that require​​ extensive manual annotation, this​​​‌ work evaluates the reuse‌ of a global plant‌​‌ identification model trained on​​ citizen science data. Using​​​‌ a vision transformer from‌ the Pl@ntNet platform, the‌​‌ study compares multi-label classification​​ and tiling-based strategies applied​​​‌ to high-resolution roadside imagery.‌ The results show that‌​‌ the tiling approach achieves​​ strong detection performance even​​​‌ without task-specific fine-tuning, demonstrating‌ the potential of large‌​‌ pretrained models for large-scale​​ invasive species monitoring at​​​‌ low cost.

From a‌ methodological perspective, 16 contributes‌​‌ to this axis by​​ proposing PlantAIM, a hybrid​​​‌ vision architecture that combines‌ global attention mechanisms with‌​‌ local feature extraction. By​​ fusing transformer-based and convolutional​​​‌ representations, the model improves‌ robustness and generalization in‌​‌ challenging plant visual recognition​​ settings, including limited training​​​‌ data and heterogeneous environments.‌ These architectural insights directly‌​‌ support the development of​​ scalable plant vision systems​​​‌ capable of reliable deployment‌ in operational monitoring scenarios.‌​‌

8.3.3 Conformal Prediction for​​​‌ uncertainty quantification

Participants: Joseph​ Salmon, Jean-Baptiste Fermanian​‌.

Deep neural networks​​ in computer vision produce​​​‌ overconfident predictions without statistical​ guarantees, making uncertainty calibration​‌ essential. Conformal prediction provides​​ distribution-free guarantees but struggles​​​‌ in the long-tailed, highly​ unbalanced settings typical of​‌ large citizen science platforms,​​ where many classes are​​​‌ rare. Recent work highlights​ both theoretical limitations and​‌ possible adaptations, including transductive​​ and grouped conformal approaches​​​‌ 42, 61.​

Handling ambiguity further requires​‌ integrating domain knowledge and​​ leveraging multiple observations of​​​‌ the same instance to​ better separate aleatoric from​‌ epistemic uncertainty. Recent conformal​​ approaches extend classification to​​​‌ multi-input settings by aggregating​ conformal p-values across observations,​‌ reducing prediction set size​​ while preserving class-conditional coverage.​​​‌ Such aggregation frameworks are​ particularly well suited to​‌ citizen science applications, where​​ multiple images per instance​​​‌ are available, and naturally​ support refined decision rules​‌ and rejection for uncertain​​ predictions 41.

8.3.4​​​‌ AI-Based Species Distribution Modeling​ and Mapping

Participants: Christophe​‌ Botella, Alexis Joly​​, Théo Larcher,​​​‌ César Leblanc, Diego​ Marcos, François Munoz​‌, Rémi Palard,​​ Lukáš Picek, Maximilien​​​‌ Servajean, Dennis Shasha​, Benjamin Bourel.​‌

This research axis focuses​​ on advancing species distribution​​​‌ modeling (SDMs) through deep​ learning and multimodal data​‌ integration, with the goal​​ of overcoming key limitations​​​‌ of classical approaches, including​ limited training data, the​‌ absence of biotic interactions,​​ and insufficient spatial resolution​​​‌ for biodiversity mapping.

A​ first line of work​‌ investigates how deep learning​​ can extend SDMs beyond​​​‌ presence-only prediction. In 14​, the team demonstrates​‌ that convolutional neural network–based​​ SDMs can effectively model​​​‌ species abundance by exploiting​ transfer learning from large​‌ presence-only datasets. This strategy​​ significantly improves abundance predictions,​​​‌ particularly for rare species​ and locally rare occurrences,​‌ and leads to clear​​ performance gains over classical​​​‌ SDMs.

A complementary direction​ explores the explicit integration​‌ of biotic structure into​​ SDMs. In 35,​​​‌ a cascading prediction framework​ is proposed in which​‌ common and dominant plant​​ species are first predicted​​​‌ from environmental variables, and​ these predictions are then​‌ used to inform the​​ distribution of less common​​​‌ species. By leveraging species​ co-occurrence patterns and competitive​‌ hierarchies, this approach improves​​ prediction accuracy at fine​​​‌ spatial resolutions, especially in​ species-rich environments.

In parallel,​‌ 46 presents a large-scale,​​ multimodal deep-SDM pipeline for​​​‌ very-high-resolution biodiversity mapping across​ Europe. Based on the​‌ integration of remote sensing​​ data, climate time series,​​​‌ and species occurrence records​ at 50 $\times$ 50​‌ m resolution, this work​​ produces continental-scale species distribution​​​‌ maps, biodiversity indicators, and​ habitat maps. The approach​‌ enables joint modeling of​​ interspecies dependencies and large-scale​​​‌ inference from heterogeneous data​ sources, supporting operational biodiversity​‌ monitoring at unprecedented spatial​​ detail.

8.3.5 Coral Reef​​​‌ Monitoring

Participants: Matteo Contini​, Sylvain Bonhommeau,​‌ Victor Illien, Sylvain​​ Poulain, Serge Bernard​​​‌, Julien Barde,​ Alexis Joly.

This​‌ research axis, conducted in​​ close collaboration with Ifremer​​​‌ and IRD, develops scalable​ AI-based methods for coral​‌ reef monitoring by combining​​ citizen-driven data collection, multi-scale​​ imaging, and deep learning.​​​‌ The overarching objective is‌ to enable accurate, fine-grained‌​‌ ecological assessment over large​​ reef areas while relying​​​‌ on low-cost and operational‌ data acquisition.

The Seatizen‌​‌ Atlas 18 provides the​​ data backbone of this​​​‌ effort, bringing together more‌ than 1.6 million underwater‌​‌ and aerial images collected​​ in shallow tropical environments​​​‌ by citizens and researchers‌ using diverse platforms. The‌​‌ dataset captures the strong​​ variability inherent to real-world​​​‌ marine imagery and is‌ distributed through an open,‌​‌ standards-compliant workflow, enabling large-scale​​ training and reuse of​​​‌ AI models for marine‌ biodiversity mapping.

Building on‌​‌ this resource, the collaboration​​ explores how fine-scale ecological​​​‌ information extracted from underwater‌ imagery can be transferred‌​‌ to broader spatial scales.​​ In 17, a​​​‌ multi-scale learning framework propagates‌ detailed coral and habitat‌​‌ classifications from underwater images​​ to drone-based aerial imagery​​​‌ through knowledge distillation. This‌ approach is further extended‌​‌ in 39, which​​ introduces a weakly supervised​​​‌ semantic segmentation method that‌ combines underwater-derived supervision, spatial‌​‌ interpolation, and self-distillation to​​ minimize annotation effort. Together,​​​‌ these contributions demonstrate how‌ multi-scale deep learning and‌​‌ weak supervision can support​​ cost-effective, high-resolution coral reef​​​‌ monitoring at scale.

8.3.6‌ Evaluation of Species Identification‌​‌ and Prediction Algorithms

Participants:​​ Alexis Joly, Lukáš​​​‌ Picek, Hervé Goëau‌, Christophe Botella,‌​‌ Diego Marcos, César​​ Leblanc, Théo Larcher​​​‌.

This research axis‌ focuses on the large-scale,‌​‌ rigorous evaluation of species​​ identification and prediction algorithms,​​​‌ with the objective of‌ characterizing state-of-the-art performance under‌​‌ realistic conditions and identifying​​ key methodological challenges for​​​‌ biodiversity-oriented AI systems.

A‌ central activity in this‌​‌ area is the organization​​ of the LifeCLEF evaluation​​​‌ campaign 45, 57‌, which continues to‌​‌ attract hundreds of research​​ teams and data scientists​​​‌ worldwide. The 2025 edition‌ featured five complementary, data-driven‌​‌ tasks covering a wide​​ range of ecological modalities​​​‌ and problem settings: AnimalCLEF‌ for open-set individual animal‌​‌ re-identification, BirdCLEF+ for species​​ recognition in complex acoustic​​​‌ soundscapes, FungiCLEF for few-shot‌ classification of rare species,‌​‌ GeoLifeCLEF 51 for plant​​ species distribution prediction from​​​‌ multimodal environmental data, and‌ PlantCLEF 49 for identifying‌​‌ multiple co-occurring plant species​​ in vegetation-plot imagery. Together,​​​‌ these benchmarks provide a‌ unique and controlled view‌​‌ of current capabilities and​​ limitations in species-level AI.​​​‌

A key insight emerging‌ across tasks is the‌​‌ persistent impact of domain​​ shift, particularly when training​​​‌ and test data differ‌ in geography, sensing modality,‌​‌ or species composition. While​​ baseline models offered strong​​​‌ starting points, the most‌ effective solutions relied on‌​‌ large-scale pretraining, self-supervised and​​ semi-supervised learning, and multimodal​​​‌ data fusion. The results‌ of BirdCLEF+ highlighted the‌​‌ potential of unlabeled audio​​ data through contrastive learning,​​​‌ whereas GeoLifeCLEF exposed the‌ difficulty of generalizing even‌​‌ with high-resolution, multimodal inputs.​​ Similarly, FungiCLEF and PlantCLEF​​​‌ confirmed that few-shot and‌ weakly supervised scenarios remain‌​‌ challenging, despite progress enabled​​ by vision transformers, prototype-based​​​‌ methods, and metadata-aware pipelines.‌ Overall, multimodality consistently emerged‌​‌ as a key driver​​ of robustness and performance,​​​‌ alongside growing interest in‌ efficient and deployable architectures.‌​‌

Complementing these benchmarking efforts,​​​‌ 29 explores species identification​ in a heritage biodiversity​‌ context, focusing on herbarium​​ specimens. This study compares​​​‌ hyperspectral leaf reflectance measurements​ with RGB image-based identification​‌ using Pl@ntNet, showing that​​ spectral approaches can achieve​​​‌ high species-level accuracy from​ relatively small datasets, even​‌ in the absence of​​ reproductive structures. The results​​​‌ highlight the complementarity of​ spectral and vision-based methods​‌ and point to practical​​ solutions for reducing taxonomic​​​‌ knowledge gaps in large​ digitized collections.

8.3.7 Importance​‌ of fossil pollen data​​ for vegetation species distribution​​​‌ modeling

Participants: Benjamin Bourel​, Christophe Botella.​‌

Given the current acceleration​​ of climate change, anticipating​​​‌ future responses in plant​ biodiversity is a major​‌ scientific and societal challenge.​​ We propose a resolutely​​​‌ innovative approach to improve​ the predictability of European​‌ vegetation dynamics, based on​​ a long-term perspective covering​​​‌ the last 20,000 years.​ By combining, for the​‌ first time on a​​ European scale, more than​​​‌ 72,000 harmonized fossil pollen​ records, high-resolution paleoclimate simulations​‌ and indicators of anthropogenic​​ pressure, we aim to​​​‌ unravel the respective roles​ of climate and human​‌ activities in past ecosystem​​ transformations. We are tackling​​​‌ a major conceptual barrier​ in ecology and paleoecology:​‌ the validity of the​​ principle of actualism and​​​‌ the underestimation of plant​ species' climatic niches due​‌ to niche truncation.

The​​ preliminary results obtained in​​​‌ 2025 during a Master's​ internship carried out by​‌ Marion Cann and supervised​​ by Benjamin Bourel and​​​‌ Christophe Botella, researchers who​ will supervise this postdoctoral​‌ project, are very encouraging.​​ The coupling of LegacyPollen​​​‌ 1.0 data with the​ paleoclimate simulations highlighted that​‌ the climatic hypervolume occupied​​ by Olea in Europe​​​‌ increased by 25% when​ taking into account past​‌ data (data in Europe​​ since the Last Glacial​​​‌ Maximum), in addition to​ present data. The integration​‌ of fossil data therefore​​ makes it possible to​​​‌ identify plant communities with​ no modern analogues and​‌ to reconstruct more realistic​​ fundamental niches for key​​​‌ taxa in European ecosystems.​ These methodological advances pave​‌ the way for more​​ robust species distribution models,​​​‌ capable of improving biodiversity​ projections in the face​‌ of future climate change.​​

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

10.1.2‌ Participation in other International‌​‌ Programs

IVADO-Inria Program:​​ IROKO & University of​​​‌ Montreal have been selected‌ as one of the‌​‌ 8 projects funded within​​ the 2025 edition of​​​‌ the IVADO-Inria Program.‌ Alexis Joly visited the‌​‌ IRBV lab in Montreal​​ two weeks in October​​​‌ and Etienne Laliberté visited‌ IROKO in Montpellier two‌​‌ weeks in November. These​​ exchanges have helped consolidate​​​‌ and structure the scientific‌ collaboration already underway around‌​‌ the interface between artificial​​ intelligence and plant ecology,​​​‌ particularly with regard to‌ the challenges of rapid‌​‌ monitoring of plant biodiversity​​ using drones. It has​​​‌ strengthened exchanges between teams‌ at the University of‌​‌ Montreal (Department of Biological​​ Sciences, IRBV, Mila) and​​​‌ the research teams leading‌ the Pl@ntNet platform (Inria‌​‌ IROKO, UMR AMAP).

10.2.1​​​‌ Visits of international scientists‌

10.3.1 Horizon Europe​​

PARAD (PARSADA),​ (2025-2030), 7.7 MEuros.

Participants:​‌ Benjamin Bourel, Alexis​​ Joly, Thomas Paillot​​.

The cross-disciplinary PARAD​​​‌ project aims to anticipate,‌ innovate and support the‌​‌ agroecological transition in weed​​ management by overcoming the​​​‌ obstacles created by the‌ reduction in herbicides and‌​‌ the withdrawal of molecules​​ through (i) a better​​​‌ understanding of the biological‌ characteristics (traits) of weeds‌​‌ that are identified as​​ being responsible for the​​​‌ failure of practices or‌ the conditions that allow‌​‌ species to circumvent practices​​ (WP1), (ii) quantifying/optimizing existing​​​‌ agroecological levers (WP2), (iii)‌ promoting technical and technological‌​‌ innovation in order to​​ detect, identify and manage​​​‌ weeds using alternative methods‌ to herbicides (WP3), (iv)‌​‌ quantitative analyses through simulations​​ and field trials of​​​‌ weed management effectiveness from‌ a multi-criteria assessment (crop‌​‌ yield, GHGs (greenhouse gases)​​ emissions,

impact on biodiversity,​​​‌ etc.) (WP4), (v) support‌ for the collective design‌​‌ of systemic solutions, through​​ case studies involving farmers​​​‌ and other local stakeholders‌ (WP5), (vi) a renewed‌​‌ interest in the recognition​​ and biology of weeds​​​‌ in order to take‌ the right action through‌​‌ initial and ongoing training​​ (WP6).

Led by INRAE,​​​‌ PARAD brings together 19‌ funded partners and 146‌​‌ permanent staff from these​​ organizations. For practical reasons,​​​‌ Iroko did not wish‌ to participate directly in‌​‌ this project as a​​ partner. However, Iroko is​​​‌ fully involved in WP3.‌ Benjamin Bourel is Pl@ntNet‌​‌ advisor of WP3. Iroko​​ is involved in defining​​​‌ standardized protocols for data‌ acquisition via smartphone and‌​‌ equivalent devices, on a​​ 1m² plot scale. It​​​‌ is also involved in‌ defining annotation formats and‌​‌ metadata. The aim is​​ to ensure the effective​​​‌ integration, storage and reuse‌ of this data within‌​‌ INRIA's Pl@ntNet ecosystem. This​​ will enable the project​​​‌ to take full advantage‌ of existing infrastructure, tools‌​‌ and communities. To this​​ end, the Pl@ntNet API​​​‌ and batch import tools‌ are being developed to‌​‌ explicitly support plot-related data.​​ This is a mutually​​​‌ beneficial relationship for Iroko‌ and PARAD. PARAD benefits‌​‌ from Pl@ntNet's experience and​​ infrastructure for plant identification.​​​‌ For its part, Pl@ntNet‌ benefits from the project's‌​‌ numerous partners, infrastructure and​​ resources to acquire large​​​‌ quantities of plant images‌ labeled by professionals and‌​‌ to test these APIs​​ and the new Pl@ntNet​​​‌ features (notably the one‌ for multi-species identification).

Past2ECO‌​‌ (PEPR Agroécologie et Numérique),​​ (2026-2031), 3 MEuros.

Participants:​​​‌ Benjamin Bourel, Alexis‌ Joly.

Agroecology relies‌​‌ on the development of​​ new genetic diversity (even​​​‌ lost ones) and practices‌ (including varietal mixture) ensuring‌​‌ ecosystemic services (beyond yield​​ stability) with limited to​​​‌ zero inputs (fertilizers and‌ pesticides) and facing climatic‌​‌ variability and extreme events​​ in the context of​​​‌ ongoing climate change. Past2ECO‌ proposes to investigate both‌​‌ genetic diversity and agricultural​​ practices relevant to agroecology​​​‌ by combining between- and‌ within-crop past (exploiting herbarium‌​‌ specimens) and contemporaneous (leveraging​​ seed banks) genetic diversity​​​‌ of wheat and sorghum,‌ to provide practical and‌​‌ knowledge-driven solutions for climate-resilient​​ agriculture and support the​​​‌ agroecological transition.

Past2ECO brings‌ together complementary expertise in‌​‌ botany, genomics, biology, agronomy,​​ AI and computer vision​​​‌ from eight institutes in‌ integrating historical genetic knowledge,‌​‌ cutting-edge genomics, and AI-based​​​‌ phenotyping tools to guide​ agroecological crop transitions. Past2ECO​‌ aims to decipher historic​​ diversity (WP1), unveil adaptive​​​‌ genomic footprints (WP2) evaluated​ in the field (WP3),​‌ all using AI-based image​​ analysis technologies (WP4). The​​​‌ WP4 of the project​ is led by Iroko​‌ (Benjamin Bourel).

Triticeae and​​ sorghum herbarium collections available​​​‌ are estimated to contain​ 18 101 specimens from​‌ 114 countries spanning 321​​ years from 1700-2021. After​​​‌ curation and classification, using​ AI-based morphometry, the analysis​‌ of ancient DNA, compared​​ to that of worldwide​​​‌ accessions hosted in seed​ banks, will allow us​‌ to document the genetic​​ diversity and the evolutionary​​​‌ trajectory of the use​ of varietal mixtures from​‌ past to present-day, and​​ assess changes in their​​​‌ soil-root metagenome associations. Exploiting​ diversity, from the past​‌ to the present using​​ innovative genomic offset statistics,​​​‌ we will be able​ to predict the optimal​‌ genotypes and varietal mixtures​​ for current and future​​​‌ climate. We will validate​ how such predicted adaptive​‌ diversity manifests phenotypically in​​ the field, how they​​​‌ can be used to​ guide the development of​‌ agroecological practices such as​​ varietal mixtures in a​​​‌ climatic gradient, and what​ their benefits and evolutionary​‌ dynamics are under realistic​​ on-farm diversity management practices.​​​‌

Past2ECO builds a strong​ bridge between computer science​‌ and agroecology in unveiling​​ specimen classification from AI-based​​​‌ geometric morphometrics, the development​ of new machine learning​‌ methods using neural networks​​ on hyperspectral images to​​​‌ discriminate between varieties, and​ digital leaf phenotypes from​‌ herbarium specimens as an​​ indicator of adaptation to​​​‌ global climate change.

Overall,​ Past2ECO will contribute to​‌ PEPR ‘Agroecology and ICT’​​ by delivering a cutting-edge​​​‌ proof-of-concept project aiming to​ decipher and exploit past​‌ (so-called lost or underused)​​ adaptive diversity of current​​​‌ and future major crop​ species, wheat and sorghum,​‌ to design varieties for​​ agroecological transitions in the​​​‌ context of climate change.​

Pl@ntAgroEco (PEPR Agroécologie et​‌ Numérique), (2023-2027), 1.6 MEuros.​​

Participants: Antoine Affouard,​​​‌ Christophe Botella, Hervé​ Goëau, Hugo Gresse​‌, Alexis Joly,​​ Thomas Paillot.

Agroecology​​​‌ necessarily involves crop diversification,​ but also the early​‌ detection of diseases, deficiencies​​ and stresses (hydric, etc.),​​​‌ as well as better​ management of biodiversity. The​‌ main stumbling block is​​ that this paradigm shift​​​‌ in agricultural practices requires​ expert skills in botany,​‌ plant pathology and ecology​​ that are not generally​​​‌ available to those working​ in the field, such​‌ as farmers or agri-food​​ technicians. Digital technologies, and​​​‌ artificial intelligence in particular,​ can play a crucial​‌ role in removing this​​ barrier to access to​​​‌ knowledge.

The aim of​ the Pl@ntAgroEco project will​‌ be to design, experiment​​ with and develop new​​​‌ high-impact agro-ecology services within​ the Pl@ntNet platform. This​‌ includes : AI and​​ plant science research; agile​​​‌ development of new components​ within the platform; organizing​‌ participatory science programs and​​ animating the Pl@ntNet user​​​‌ community. The project is​ leaded by Iroko (Alexis​‌ Joly).

FishPredict (ANR), (2022-2025),​​ 500 KEuros.

Participants: Benjamin​​​‌ Bourel, Alexis Joly​, Maximilien Servajean,​‌ Julien Thomazo.

FishPredict​​ ANR project funded in​​ the context of the​​​‌ IA-Biodiv challenge. The projects‌ aims at predicting the‌​‌ biodiversity of reef fishes​​ using AI technologies. Alexis​​​‌ Joly is co-leading of‌ the whole project jointly‌​‌ with David Mouillot, marine​​ ecologist at the MARBEC​​​‌ lab.

DeepPEP (ANR), (2025-2027),‌ 25 KEuros.

Participants: Reza‌​‌ Akbarinia, Dennis Shasha​​, Patrick Valduriez.​​​‌

The DeepPEP project, between‌ CNRS, INRAE and Inria‌​‌ Iroko, aims to enhance​​ the fundamental understanding of​​​‌ nutrient homeostasis in plants‌ and develop new biostimulants‌​‌ using signaling peptides. The​​ main objective is to​​​‌ acquire fundamental knowledge on‌ the control of nutrient‌​‌ homeostasis in plants and​​ to develop new biotechnological​​​‌ resources in the form‌ of signaling peptides as‌​‌ biostimulants​​. The project seeks​​ to create new AI​​​‌ algorithms for designing peptides‌ that interact with any‌​‌ protein and develop potential​​ biostimulants to enhance nitrogen​​​‌ (N) and phosphorus (P)‌ efficiency in agriculture​​. In‌​‌ this project, Iroko provides​​ its expertise in time​​​‌ series query processing

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

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​‌ analyzing antimicrobial resistance (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.

PNR "Beerisk" (2022-2025).​​ 200 KEuros.

Participants: Reza​​​‌ Akbarinia, Florent Masseglia‌.

The objective of‌​‌ this project is to​​ analyze honeybee daily mortality​​​‌ rates, represented as time‌ series, in order to‌​‌ detect anomalies and study​​ the lethal effects of​​​‌ bees exposure to pesticides.‌

Plan national Ecoantibio "INTERSECTION"‌​‌ (2024-2028), 175 Keuros

Participants:​​ Reza Akbarinia, Florent​​​‌ Masseglia.

The objective‌ of the INTERSECTION project‌​‌ is to produce intersectoral​​ and territorial indicators for​​​‌ monitoring resistance and use‌ of antibiotics in France,‌​‌ and to facilitate the​​ use and analysis of​​​‌ these indicators, in a‌ One health approach.

PEPR‌​‌ agroécologie et numérique "RootSystemTracker"​​ (2024-2027), 144 Keuros

Participants:​​​‌ Reza Akbarinia, Christophe‌ Pradal, Lo'Ai Gandeel‌​‌.

Roots play a​​ crucial role in nutrient​​​‌ and water uptake, atmospheric‌ carbon fixation, and soil‌​‌ interactions, significantly influencing resource​​ use efficiency and crop​​​‌ resilience to environmental stresses.‌ The objective of the‌​‌ RootSystemTracker project is to​​ develop efficient methods for​​​‌ the spatio-temporal phenotyping of‌ plant root architectures using‌​‌ heterogeneous data. This involves​​ automatically capturing their topology​​​‌ and geometry over time,‌ despite challenges such as‌​‌ root occlusions and variability​​ in observation conditions.

Inria​​​‌ Challenge OMICFINDER (2023-2027), 1‌ Engineer - 24 months‌​‌

Participants: Reza Akbarinia,​​ Rebecca Pontes Salles,​​​‌ Florent Masseglia.

While‌ genomic sequencing is enabling‌​‌ crucial advances in medicine,​​ ecology, and agriculture, the​​​‌ exponential growth of public‌ databases (48 petabytes by‌​‌ 2023) remains largely untapped​​​‌ due to the lack​ of efficient querying methods.​‌ OMICFINDER proposes an innovative​​ global search engine that​​​‌ makes it possible to​ query nucleotidic sequences against​‌ the vast amount of​​ publicly available genomic data.​​​‌ Combining novel algorithms, semantic​ web technologies, and distributed​‌ indexing with a focus​​ on environmental sustainability, it​​​‌ aims to unlock this​ treasure trove of information​‌ – bringing the equivalent​​ of a search engine​​​‌ to genomics at last.​ The project is led​‌ by Pierre Peterlongo (GenScale​​ team, Inria Rennes).

10.4.1​​​‌ Others

Participants: Alexis Joly​, Jean-Christophe Lombardo,​‌ Hervé Goëau, Hugo​​ Gresse, Mathias Chouet​​​‌, Antoine Affouard,​ David Margery.

Pl@ntNet​‌ consortium: In 2025, CNRS​​ has joined the Pl@ntNet​​​‌ consortium as a new​ member. This contract, initially​‌ signed by four founding​​ research organisms (Inria, CIRAD,​​​‌ IRD, INRAE) aims at​ sustaining the Pl@ntNet platform​‌ in the long term.​​ It has been initiated​​​‌ in November 2019 in​ the context of the​‌ InriaSOFT national program of​​ Inria. Each partner subscribes​​​‌ a yearly subscription (10-20K​ euros per year) to​‌ cover engineering costs for​​ maintenance and technological developments.​​​‌ Depending on the membership​ status, each partner has​‌ one vote in the​​ steering committee and/or the​​​‌ technical committee of the​ platform. He can also​‌ use the platform in​​ his own projects and​​​‌ benefit from a certain​ number of service days​‌ within the platform. The​​ consortium is not fixed​​​‌ and is intended to​ be extended to other​‌ members in the coming​​ years.

Regional project "DACLIM" (2023-2026),​ 70 Keuros

Participants: Reza​‌ Akbarinia, Florent Masseglia​​, Guillaume Coulaud.​​​‌

The objective of this​ project is to develop​‌ scalable techniques based on​​ massive data distribution to​​​‌ enable the efficient detection​ of anomalies in large​‌ climate databases. The detection​​ of anomalies in climate​​​‌ data can provide climatologists​ with insights into the​‌ behavior of various climatological​​ variables, understanding of extreme​​​‌ events such as heatwaves​ and cold snaps, as​‌ well as the prediction​​ of these types of​​​‌ events.

CESE consultation on​‌ the impact of AI​​ on the environment

The​​​‌ CESE (Conseil Economique, Social​ et Environnemental) is one​‌ of the 3 assemblies​​ of the French constitution,​​​‌ made up of elected​ representatives of civil society​‌ (unions, associations, companies, students,​​ etc.). Its role is​​​‌ to provide advice on​ economic, social and environmental​‌ policies to guide public​​ decision-making (governmental in particular).​​​‌ Alexis Joly took part​ in the consultation entitled​‌ “Impacts of artificial intelligence:​​ risks and opportunities for​​​‌ the environment”. He was​ consulted and interviewed on​‌ several occasions and was​​ one of the 3​​​‌ experts invited to the​ final plenary session that​‌ voted on the recommendations.​​

OECD report on the​​​‌ advancement of the productivity​ of science with citizen​‌ science and artificial intelligence​​

Alexis Joly is a​​​‌ co-author of the chapter​ “Advancing the productivity of​‌ science with citizen science​​ and artificial intelligence” in​​​‌ the OECD report Artificial​ Intelligence in Science: Challenges,​‌ Opportunities and the Future​​ of Research (PDF​​). He participated in​​​‌ all preparatory meetings and‌ contributed approximately 15% of‌​‌ the chapter, based on​​ Iroko’s expertise in citizen​​​‌ science, large-scale ecological data,‌ and AI-driven biodiversity monitoring.‌​‌

11.1.1 Scientific​​​‌ events: organization

11.1.2 Scientific events:‌ selection

11.1.3​​ Journal

11.1.4 Invited talks

• Patrick‌​‌ Valduriez gave an invited​​ talk on AI and​​​‌ Scientific Research on:
  • July‌ 2, 2025 at Cirad,‌​‌ Montpellier;
  • October 16, 2025​​ at Inria, Montpellier;
  • July​​​‌ 17, 2025 at the‌ Dinizia workshop, CEFET-RJ, Rio‌​‌ de Janeiro, Brazil;
  • November​​ 6, 2025 at Workshop​​​‌ of the Artificial Intelligence‌ Institute, LNCC, Petropolis,‌​‌ Brazil.
• Joseph Salmon
  • invited​​ talk for IA Connect​​​‌ (launch of IA Montpellier‌ Méditerranée)
  • invited talk at‌​‌ OSCII 2025
• Alexis Joly​​
  • MILA Tea talk -​​​‌ November 17, 2025 -‌ Montreal
  • IRBV seminar -‌​‌ October 23, 2025 -​​ Montreal
  • Congrès annuel du​​​‌ Collège des Sociétés savantes‌ académiques de France -‌​‌ February 4, 2025 -​​ Montpellier
  • EU JRC workshop​​​‌ on trees and habitat‌ mapping - November 18-19,‌​‌ 2025 - Ispra

11.1.5​​ Leadership within the scientific​​​‌ community

• Esther Pacitti :‌ Member of the Steering‌​‌ Committee of the BDA​​ conference.
• Reza Akbarinia :​​​‌ Member of the Steering‌ Committee of the BDA‌​‌ conference.
• Alexis Joly
  • Scientific​​ and Technical director of​​​‌ Pl@ntNet platform
  • Coordinator of‌ the LifeCLEF international virtual‌​‌ lab
• Joseph Salmon head​​ of the Statistics and​​​‌ Data Science specialty, doctoral‌ school EDI2S

11.1.6 Scientific‌​‌ expertise

• Christophe Pradal :​​ member of the INRAE​​​‌ evaluation committee CSS (Scientific‌ Specialist Commission) in Plant‌​‌ Integrated Biology
• Reza Akbarinia​​ : member of the​​​‌ evaluation committee (section 27),‌ University of Montpellier.
• Alexis‌​‌ Joly :
  • GENCI expert​​ committee (AI thematic​​​‌)
  • Scientific Advisory Board‌ of the chaire "Angèle‌​‌ St-Pierre / Hugo Larochelle"​​ related to AI applied​​​‌ to environment
  • Expert for‌ SNSF (Swiss National Science‌​‌ Foundation) - project scientific​​ evaluation
• Joseph Salmon :​​​‌
  • elected member of the‌ "Commission de Section 26",‌​‌ Univ. Montpellier.
  • head of​​ the jury for hiring​​​‌ assistant professor (Univ. Montpellier,‌ Faculté des Science)
  • Member‌​‌ of the Steering Committee​​​‌ for MathPhDInFrance
• Patrick Valduriez​ : consultant on big​‌ data for the Software​​ Heritage project

11.1.7 Research​​​‌ administration

• Florent Masseglia :​ deputy scientific director of​‌ Inria for the domain​​ "Perception, Cognition and Interaction",​​​‌ 50% of his time​ until September 2025.
• Reza​‌ Akbarinia : Scientific referent​​ for research data at​​​‌ Inria branch of Montpellier;​ Member of Inria national​‌ commission for research data.​​
• Esther Pacitti : manager​​​‌ of Polytech' Montpellier's International​ Relations for the computer​‌ science department (100 students).​​
• Patrick Valduriez : scientific​​​‌ manager for the Latin​ America zone at Inria's​‌ Direction of Foreign Relationships​​ (DRI) and scientific director​​​‌ of the Inria-Brasil strategic​ partnership.
• Christophe Pradal :​‌ Team leader with C.​​ Granier of the PhenoMEn​​​‌ team of the AGAP​ Institute.
• Alexis Joly :​‌ co-manager of a Collaborative​​ Doctoral Partnership between the​​​‌ EU Joint Research Centre​ of Ispra and the​‌ University of Montpellier

11.2.1 Teaching​‌

Esther Pacitti :

• 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.​​

Joseph Salmon :

• HAX603X:​​​‌ Stochastic Modeling, 20h, level​ L3, 50 students.
• Supervision​‌ of master internships.
• Supervision​​ of data science discovery​​​‌ projects.

11.2.2 Supervision

PhD​ & HDR:

• PhD (defended):​‌ Cesar Leblanc, Predicting biodiversity​​ future trajectories through deep​​​‌ learning. Advisors: Alexis Joly​ , Maximilien Servajean, Pierre​‌ Bonnet.
• PhD (defended 58​​): Matteo Contini, Multi-scale​​​‌ monitoring of coastal marine​ biodiversity. Advisors: Sylvain Bonhommeau,​‌ Alexis Joly .
• PhD​​ in progress: Kawtar Zaher,​​​‌ Novel class retrieval through​ interactive learning. Advisors: Olivier​‌ Buisson, Alexis Joly .​​
• PhD in progress: Guillaume​​​‌ Coulaud, Anomaly Detection in​ Big Climate Data. Advisors:​‌ Reza Akbarinia , Audrey​​ Brouillet, Florent Masseglia .​​​‌
• PhD in progress: Loaï​ Gandeel, Automatic methods for​‌ spatio-temporal reconstruction of root​​ architecture. Advisors: Reza Akbarinia​​​‌ , Romain Fernandez, Christophe​ Pradal .
• PhD in​‌ progress: Raphaël Benerradi, species​​ trends estimation from citizen​​​‌ science data. Advisors: Christophe​ Botella , Alexis Joly​‌ , Maximilien Servajean.
• PhD​​ in progress: Théo Larcher,​​​‌ multi-scale species prediction. Advisors:​ Alexis Joly , Joseph​‌ Salmon , Pierre Bonnet,​​ Marijn Van der Velde.​​​‌
• PhD in progress: Sébastien​ Gigot-Leandri, decision-oriented site occupancy​‌ models. Advisors: Alexis Joly​​ , Maximilien Servajean, David​​​‌ Mouillot.
• PhD in progress:​ Alex Maleknia , Influence​‌ functions and their applications​​ to machine learning. Advisors:​​​‌ Joseph Salmon , E.​ Chzhen.

11.2.3 Juries

Members​‌ of the team participated​​ in the following PhD​​​‌ or HDR committees:

• Reza​ Akbarinia :
  • Sara Jarrad,​‌ Sorbonne University (PhD reviewer)​​
  • Omar Ghannou, Aix-Marseille University​​​‌ (PhD reviewer)
• Joseph Salmon​ :
  • Benjamin Charlier (directeur​‌ du jury d'HDR, Univ.​​ Montpellier)
  • Grégoire Pacreau (rapporteur​​​‌ de la thèse, École​ Polytechnique)
• Alexis Joly :​‌
  • Cesar Leblanc PhD defense,​​ Univ. of Montpellier, (PhD​​​‌ director)
  • Matteo Contini PhD​ defense, Univ. of Montpellier​‌ (as PhD director)

11.2.4​​ Educational and pedagogical outreach​​

• Patrick Valduriez , Chiche​​​‌ (2 actions): Lycée Mermoz,‌ Lycée Clémenceau, Montpellier.
• Florent‌​‌ Masseglia , Chiche (4​​ actions): lycée Philippe de​​​‌ Girard, Avignon.

11.3.1 Specific official responsibilities‌​‌ in science outreach structures​​

• Alexis Joly :
  • Member​​​‌ of the steering committee‌ of Pl@ntNet citizen science‌​‌ platform
  • Scientific Advisory Board​​ of project Le Féral​​​‌

11.3.2 Productions (articles, videos,‌ podcasts, serious games, ...)‌​‌

• Joseph Salmon : Scientific​​ blogging
• Pl@ntNet team:
  • Pl@ntNet​​​‌ website, in particular‌ the news section
  • Pl@ntNet‌​‌ documentation (new in 2025)​​
  • Videos: Participation in Xprize​​​‌, Identify a plant‌ with Pl@ntNet, PlantNet‌​‌ for dummies
  • Articles: Telabotanica​​ news, CIRAD blog​​​‌, Frandroid app of‌ the week, Gabon‌​‌ Actu.

11.3.3 Participation​​ in Live events

• Joseph​​​‌ Salmon (2 actions): Mois‌ des mathématiques appliquées et‌​‌ industrielles. (Lycée Joffre, Montpellier),​​ IAUM (Univ. Montpellier, audience​​​‌ high schoolers)

11.3.4 Others‌ science outreach relevant activities‌​‌

• Joseph Salmon : Fête​​ des sciences (carasciences), Montpellier​​​‌

International​​​‌ journals

• 14 articleB.​Benjamin Bourel, A.​‌Alexis Joly, M.​​Maximilien Servajean, S.​​​‌Simon Bettinger, J.​ A.José Antonio Sanabria-Fernández​‌ and D.David Mouillot​​. From Presence‐Only to​​​‌ Abundance Species Distribution Models​ Using Transfer Learning.​‌Ecology Letters287​​July 2025, e70177​​​‌HALDOIback to​ text
• 15 articleA.​‌Audrey Brouillet, G.​​Guillaume Coulaud, D.​​​‌Dennis Shasha, R.​Reza Akbarinia and F.​‌Florent Masseglia. ClimBurst:​​ A Novel Method to​​​‌ Detect Climatological Anomalies Over​ Time and Space.​‌Geophysical Research Letters52​​19October 2025,​​​‌ e2025GL117095HALDOIback​ to text
• 16 article​‌A. Y.Abel Yu​​ Hao Chai, S.​​ H.Sue Han Lee​​​‌, F. S.Fei‌ Siang Tay, H.‌​‌Hervé Goëau, P.​​Pierre Bonnet and A.​​​‌Alexis Joly. PlantAIM:‌ A new baseline model‌​‌ integrating global attention and​​ local features for enhanced​​​‌ plant disease identification.‌Smart Agricultural Technology10‌​‌March 2025, 100813​​HALDOIback to​​​‌ text
• 17 articleM.‌Matteo Contini, V.‌​‌Victor Illien, J.​​Julien Barde, S.​​​‌Sylvain Poulain, S.‌Serge Bernard, A.‌​‌Alexis Joly and S.​​Sylvain Bonhommeau. From​​​‌ underwater to drone: A‌ novel multi-scale knowledge distillation‌​‌ approach for coral reef​​ monitoring.Ecological Informatics​​​‌89November 2025,‌ 103149 (16p.)HALDOI‌​‌back to text
• 18​​ articleM.Matteo Contini​​​‌, V.Victor Illien‌, M.Mohan Julien‌​‌, M.Mervyn Ravitchandirane​​, V.Victor Russias​​​‌, A.Arthur Lazennec‌, T.Thomas Chevrier‌​‌, C. L.Cam​​ Ly Rintz, L.​​​‌Léanne Carpentier, P.‌Pierre Gogendeau, C.‌​‌César Leblanc, S.​​Serge Bernard, A.​​​‌Alexandre Boyer, J.‌Justine Talpaert Daudon,‌​‌ S.Sylvain Poulain,​​ J.Julien Barde,​​​‌ A.Alexis Joly,‌ S.Sylvain Bonhommeau and‌​‌ T.Thomas Chevrier.​​ Seatizen Atlas: a collaborative​​​‌ dataset of underwater and‌ aerial marine imagery.‌​‌Scientific Data 121​​2025, 67HAL​​​‌DOIback to text‌back to textback‌​‌ to text
• 19 article​​S.Sabine Demotes-Mainard,​​​‌ H.Hervé Autret,‌ C.Christophe Pradal,‌​‌ J.Julien Le Gall​​, V.Vincent Guérin​​​‌, N.Nathalie Leduc‌, D.Didier Combes‌​‌, C.Christophe Renaud​​, M.Michaël Chelle​​​‌ and J.Jessica Bertheloot‌. Simulating light quantity‌​‌ and quality over plant​​ organs using a ray-tracing​​​‌ method to investigate plant‌ responses in growth chambers‌​‌.Biosystems Engineering258​​September 2025, 104256​​​‌HALDOI
• 20 article‌V.Vincent Espitalier,‌​‌ J.-C.Jean-Christophe Lombardo,​​ H.Hervé Goëau,​​​‌ C.Christophe Botella,‌ T. T.Toke Thomas‌​‌ Høye, M.Mads​​ Dyrmann, P.Pierre​​​‌ Bonnet and A.Alexis‌ Joly. Adapting a‌​‌ global plant identification model​​ to detect invasive alien​​​‌ plant species in high-resolution‌ road side images.‌​‌Ecological Informatics89May​​ 2025, 103129HAL​​​‌DOIback to text‌
• 21 articleS.Sabrina‌​‌ Kumschick, L.Lysandre​​ Journiac, O.Océane​​​‌ Boulesnane-Genguant, C.Christophe‌ Botella, R.Robin‌​‌ Pouteau and M.Mathieu​​ Rouget. Mapping potential​​​‌ environmental impacts of alien‌ species in the face‌​‌ of climate change.​​Biological Invasions271​​​‌January 2025, 43‌HALDOI
• 22 article‌​‌L.Lorraine Latchoumane,​​ M.Martin Ecarnot,​​​‌ R.Ryad Bendoula,‌ J.-M. J.Jean-Michel J.‌​‌ -M. Roger, S.​​Sílvia Mas Garcia,​​​‌ H.Héloïse Villesseche,‌ F.Flora Tavernier,‌​‌ M.Maxime Ryckewaert,​​ N. J.Nathalie J.​​​‌ B. Gorretta, P.‌Pierre Roumet and E.‌​‌Elsa Ballini. Early​​ detection of Zymoseptoria tritici​​​‌ infection on wheat leaves‌ using hyperspectral imaging data‌​‌.Data in Brief​​​‌59April 2025,​ 111404HALDOI
• 23​‌ articleC.César Leblanc​​, P.Pierre Bonnet​​​‌, M.Maximilien Servajean​, W.Wilfried Thuiller​‌, M.Milan Chytrý​​, S.Svetlana Aćić​​​‌, O.Olivier Argagnon​, I.Idoia Biurrun​‌, G.Gianmaria Bonari​​, H.Helge Bruelheide​​​‌, J. A.Juan​ Antonio Campos, A.​‌Andraž Čarni, R.​​Renata Ćušterevska, M.​​​‌Michele de Sanctis,​ J.Jürgen Dengler,​‌ T.Tetiana Dziuba,​​ E.Emmanuel Garbolino,​​​‌ U.Ute Jandt,​ F.Florian Jansen,​‌ J.Jonathan Lenoir,​​ J. E.Jesper Erenskjold​​​‌ Moeslund, A.Aaron​ Pérez-Haase, R.Remigiusz​‌ Pielech, J.Jozef​​ Sibik, Z.Zvjezdana​​​‌ Stančić, D.Domas​ Uogintas, T.Thomas​‌ Wohlgemuth and A.Alexis​​ Joly. Learning the​​​‌ syntax of plant assemblages​.Nature Plants11​‌October 2025, 2026–2040​​HALDOIback to​​​‌ textback to text​
• 24 articleT.Tanguy​‌ Lefort, A.Antoine​​ Affouard, B.Benjamin​​​‌ Charlier, J.-C.Jean-Christophe​ Lombardo, M.Mathias​‌ Chouet, H.Hervé​​ Goëau, J.Joseph​​​‌ Salmon, P.Pierre​ Bonnet and A.Alexis​‌ Joly. Cooperative learning​​ of Pl@ntNet's Artificial Intelligence​​​‌ algorithm: how does it​ work and how can​‌ we improve it?Methods​​ in Ecology and Evolution​​​‌February 2025. In​ press. HALDOI
• 25​‌ articleS.Sébastien Levionnois​​, N.Noémie Gaudio​​​‌, R.Rémi Mahmoud​, C.Christophe Pradal​‌ and C.Corinne Robert​​. Impact of wheat-legume​​​‌ mix intercrops on wheat​ epidemics by modelling.​‌Field Crops Research336​​February 2026, 110212​​​‌HALDOIback to​ text
• 26 articleJ.​‌Ji Liu, B.​​Beichen Ma, Q.​​​‌Qiaolin Yu, Y.​Yang Zhou, J.​‌Jingbo Zhou, R.​​Ruoming Jin, D.​​​‌Dejing Dou, H.​Huaiyu Dai, H.​‌Haixun Wang and P.​​Patrick Valduriez. Efficient​​​‌ Federated Learning with Heterogeneous​ Data and Adaptive Dropout​‌.ACM Transactions on​​ Knowledge Discovery from Data​​​‌ (TKDD)198September​ 2025, 1-31HAL​‌DOIback to text​​
• 27 article D.Diego​​​‌ Marcos, R.Robert​ van de Vlasakker,​‌ I. N.Ioannis N.​​ Athanasiadis, P.Pierre​​​‌ Bonnet, H.Hervé​ Goëau, A.Alexis​‌ Joly, W. D.​​W. Daniel Kissling,​​​‌ C.César Leblanc,​ A. S.André S.J.​‌ van Proosdij and K.​​ P.Konstantinos P. Panousis​​​‌. Fully automatic extraction​ of morphological traits from​‌ the web: Utopia or​​ reality? Applications in Plant​​​‌ Sciences 13 3 June​ 2025 HAL DOI back​‌ to text
• 28 article​​G.Giulio Martellucci,​​​‌ H.Hervé Goëau,​ P.Pierre Bonnet,​‌ F.Fabrice Vinatier and​​ A.Alexis Joly.​​​‌ PlantCLEF 2025: Advancing AI-based​ Multi-Species Plant Identification in​‌ Vegetation Quadrats for Supporting​​ Environmental Law and Biodiversity​​​‌ Monitoring.Biodiversity Information​ Science and Standards9​‌December 2025, e181733​​HALDOI
• 29 article​​​‌B. M.Barbara M.​ Neto‐bradley, P.Pierre​‌ Bonnet, H.Hervé​​ Goëau, A.Alexis​​ Joly, J.Jeannine​​​‌ Cavender‐bares and D. A.‌David A. Coomes.‌​‌ Using reflectance spectra and​​ Pl@ntNet to identify herbarium​​​‌ specimens: a case study‌ with <i>Lithocarpus</i>.New‌​‌ PhytologistJune 2025HAL​​DOIback to text​​​‌
• 30 articleE.Ellen‌ Paixão, H.Helga‌​‌ Balbi, E.Esther​​ Pacitti, F.Fabio​​​‌ Porto, J.Joel‌ Santos and E.Eduardo‌​‌ Ogasawara. FFT-Based Anomaly​​ Detectors: Cutoff Frequency Adjustment​​​‌ and SMA-Based Approach.‌Journal of Information and‌​‌ Data Management2025.​​ In press. HAL
• 31​​​‌ article L.Lukas Rauch‌, R.René Heinrich‌​‌, I.Ilyass Moummad​​, A.Alexis Joly​​​‌, B.Bernhard Sick‌ and C.Christoph Scholz‌​‌. Can Masked Autoencoders​​ Also Listen to Birds?​​​‌ Transactions on Machine Learning‌ Research Journal August 2025‌​‌ HAL
• 32 articleF.​​Francesco Reyes, B.​​​‌Benjamin Pitchers, C.‌Christophe Pradal and P.-É.‌​‌Pierre-Éric Lauri. Young​​ apple tree development under​​​‌ agroforestry radiative conditions: a‌ multi-scale morphological and architectural‌​‌ dataset.AoB Plants​​17August 2025,​​​‌ plaf029HALDOI
• 33‌ articleR.Rebecca Salles‌​‌, B.Benoit Lange​​, R.Reza Akbarinia​​​‌, F.Florent Masseglia‌, E.Eduardo Ogasawara‌​‌ and E.Esther Pacitti​​. Scalable and accurate​​​‌ online multivariate anomaly detection‌.Information Systems131‌​‌June 2025, 102524​​HALDOIback to​​​‌ text
• 34 articleR.‌ A.Rodrigo A. P.‌​‌ Silva, W.Wesley​​ Ferreira, E.Esther​​​‌ Pacitti, Y. Y.‌Yuri Y. Frota and‌​‌ D.Daniel de Oliveira​​. A Greedy Constructive​​​‌ Heuristic for Executing Cloud-based‌ Workflows with Data Confidentiality‌​‌ Restrictions.SN Computer​​ Science7January 2026​​​‌, 1-52/92HALDOI‌
• 35 articleJ.Jiexun‌​‌ Xu, K.Kimberly​​ Low, C.Christophe​​​‌ Botella, B.Benjamin‌ Deneu, D.Dennis‌​‌ Shasha and A.Alexis​​ Joly. Cascading predictions​​​‌ from common to uncommon‌ species improves species distribution‌​‌ models for plants.​​Ecological Informatics93February​​​‌ 2026, 103424HAL‌DOIback to text‌​‌

International peer-reviewed conferences

• 36​​ inproceedingsA.Ananthu Aniraj​​​‌, C. F.Cassio‌ F. Dantas, D.‌​‌Dino Ienco and D.​​Diego Marcos. PDiscoFormer:​​​‌ Relaxing Part Discovery Constraints‌ with Vision Transformers.‌​‌Proceedings of the European​​ Conference on Computer Vision​​​‌ (ECCV)ECCV 2024 -‌ 18th European Conference on‌​‌ Computer Vision15143Lecture​​ Notes in Computer Science​​​‌Milano, ItalySpringer Nature‌ SwitzerlandNovember 2025,‌​‌ 256-272HALDOI
• 37​​ inproceedingsJ. S.Juan​​​‌ Sebastián Cañas, S.‌Stefan Kahl, T.‌​‌Tom Denton, M.​​ P.María Paula Toro-Gómez​​​‌, S.Susana Rodriguez-Buritica‌, J. L.Jose‌​‌ Luis Benavides-Lopez, J.​​ S.Juan Sebastián Ulloa​​​‌, P.Paula Caycedo-Rosales‌, H.Holger Klinck‌​‌, H.Hervé Goëau​​, R.Robert Planqué​​​‌, W.-P.Willem-Pier Vellinga‌ and A.Alexis Joly‌​‌. Overview of BirdCLEF+​​ 2025: Multi-Taxonomic Sound Identification​​​‌ in the Middle Magdalena,‌ Colombia.CEUR workshop‌​‌CLEF 2025 - Working​​ Notes of the Conference​​​‌ and Labs of the‌ Evaluation ForumCEUR-4038CEUR‌​‌ Workshop Proceedings232Madrid,​​​‌ SpainCEUR2025,​ 2909-2919HAL
• 38 inproceedings​‌A. Y.Abel Yu​​ Hao Chai, K.​​​‌ L.Kelly Li Zhen​ Jee, S. H.​‌Sue Han Lee,​​ F. S.Fei Siang​​​‌ Tay, J.Jules​ Vandeputte, H.Hervé​‌ Goeau, P.Pierre​​ Bonnet and A.Alexis​​​‌ Joly. Deep-Plant-Disease Dataset​ Is All You Need​‌ for Plant Disease Identification​​.ACM digital library​​​‌ACM MM 2025 -​ 33. ACM International Conference​‌ on MultimediaMM '25:​​ Proceedings of the 33rd​​​‌ ACM International Conference on​ MultimediaDublin, IrelandACM​‌October 2025, 12578-12584​​HALDOIback to​​​‌ text
• 39 inproceedingsM.​Matteo Contini, V.​‌Victor Illien, S.​​Sylvain Poulain, S.​​​‌Serge Bernard, J.​Julien Barde, S.​‌Sylvain Bonhommeau and A.​​Alexis Joly. The​​​‌ point is the mask:​ scaling coral reef segmentation​‌ with weak supervision.​​ICCV 2025 - Joint​​​‌ Workshop on Marine Vision​Honolulu, Hawaii, United States​‌October 2025HALback​​ to text
• 40 inproceedings​​​‌G.Guillaume Coulaud,​ B.Benoit Lange,​‌ D.Dennis Shasha,​​ A.Audrey Brouillet,​​​‌ R.Reza Akbarinia and​ F.Florent Masseglia.​‌ ClimBurst: A Dynamic Visualization​​ Tool to Display Climatological​​​‌ Anomalies over Time and​ Space.Proceedings of​‌ the 34th ACM International​​ Conference on Information and​​​‌ Knowledge ManagementCIKM 2025​ - 34th ACM International​‌ Conference on Information and​​ Knowledge ManagementSeoul, South​​​‌ KoreaACMNovember 2025​, 6629-6633HALDOI​‌back to text
• 41​​ inproceedingsJ.-B.Jean-Baptiste Fermanian​​​‌, M.Mohamed Hebiri​ and J.Joseph Salmon​‌. Class conditional conformal​​ prediction for multiple inputs​​​‌ by p-value aggregation.​NeurIPS 2025, The Thirty-Ninth​‌ Annual Conference on Neural​​ Information Processing SystemsSan​​​‌ Diego (CA), United States​December 2025HALback​‌ to text
• 42 inproceedings​​J.-B.Jean-Baptiste Fermanian,​​​‌ P.Pierre Humbert and​ G.Gilles Blanchard.​‌ Transductive Conformal Inference for​​ Full Ranking.NeurIPS​​​‌ 2025, The Thirty-Ninth Annual​ Conference on Neural Information​‌ Processing SystemsSan Diego​​ (CA), United StatesDecember​​​‌ 2025HALback to​ text
• 43 inproceedingsC.​‌Camille Garcin, M.​​Maximilien Servajean, A.​​​‌Alexis Joly and J.​Joseph Salmon. A​‌ two-head loss function for​​ deep Average-K classification.​​​‌WACV 2025 - IEEE/CVF​ Winter Conference on Applications​‌ of Computer VisionTucson,​​ United StatesIEEE2025​​​‌, 7358-7367HALDOI​
• 44 inproceedingsL.Lucas​‌ Giusti Tavares, J.​​Janio Lima, M.​​​‌Matheus Melo, C.​Chao Chen, J.​‌ M.Jonathan M. Garibaldi​​, G.Gabriel dos​​​‌ Santos Scatena, A.​ H.Anna Helena Reali​‌ Costa, E. S.​​Edson Satoshi Gomi,​​​‌ R.Rebecca Salles,​ E.Esther Pacitti,​‌ I.Ismael Santos,​​ I. G.Isabela Guimarães​​​‌ Siqueira, F.Fabio​ Porto, D.Diego​‌ Carvalho, R.Rafaelli​​ Coutinho and E.Eduardo​​​‌ Ogasawara. Fuzzy-Based Ensemble​ Method for Robust Concept​‌ Drift Detection in Multivariate​​ Time Series.IEEE​​​‌ XploreIJCNN 2025 -​ International Joint Conference on​‌ Neural NetworksRome, Italy​​IEEE2025, 1-12​​HALDOIback to​​​‌ text
• 45 inproceedingsA.‌Alexis Joly, L.‌​‌Lukáš Picek, S.​​Stefan Kahl, H.​​​‌Hervé Goëau, L.‌Lukáš Adam, C.‌​‌Christophe Botella, M.​​Maximilien Servajean, D.​​​‌Diego Marcos, C.‌Cesar Leblanc, T.‌​‌Théo Larcher, J.​​Jiří Matas, K.​​​‌Klára Janoušková, V.‌Vojtěch Čermák, K.‌​‌Kostas Papafitsoros, R.​​Robert Planqué, W.-P.​​​‌Willem-Pier Vellinga, H.‌Holger Klinck, T.‌​‌Tom Denton, P.​​Pierre Bonnet and H.​​​‌Henning Müller. LifeCLEF‌ 2025 Teaser: Challenges on‌​‌ Species Presence Prediction and​​ Identification, and Individual Animal​​​‌ Identification.Lecture notes‌ in computer science47.‌​‌ European Conference on Information​​ Retrieval - ECIR 2025​​​‌Advances in Information Retrieval.‌ ECIR 2025 : part‌​‌ VLNCS-15576Lucca, Italy​​Springer Nature SwitzerlandApril​​​‌ 2025, 373-381HAL‌DOIback to text‌​‌
• 46 inproceedingsC.César​​ Leblanc, L.Lukáš​​​‌ Picek, B.Benjamin‌ Deneu, P.Pierre‌​‌ Bonnet, M.Maximilien​​ Servajean, R.Rémi​​​‌ Palard and A.Alexis‌ Joly. Mapping Biodiversity‌​‌ at Very-High Resolution in​​ Europe.CVPR 2025​​​‌ - IEEE/CVF Computer Vision‌ and Pattern Recognition Conference‌​‌Nashville, TN, United States​​IEEEJune 2025,​​​‌ 2340-2349HALDOIback‌ to text
• 47 inproceedings‌​‌J. Z.Jerad Zherui​​ Liaw, A. Y.​​​‌Abel Yu Hao Chai‌, S. H.Sue‌​‌ Han Lee, P.​​Pierre Bonnet and A.​​​‌Alexis Joly. Can‌ Language Improve Visual Features‌​‌ For Distinguishing Unseen Plant​​ Diseases?ICPR 2024 -​​​‌ 27th International Conference on‌ Pattern Recognition15330Lecture‌​‌ Notes in Computer Science​​Kolkata, IndiaSpringer Nature​​​‌ SwitzerlandJanuary 2025,‌ 296-311HALDOI
• 48‌​‌ inproceedingsJ.Janio Lima​​, H.Hélio Castro​​​‌, L.Luiz Oliveira‌, E.Ellen Paixão‌​‌, L.Lais Baroni​​, R.Rebecca Salles​​​‌, R.Ricardo Vargas‌ and E.Eduardo Ogasawara‌​‌. UniTED: A Unified​​ Time Series Event Detection​​​‌ Repository.Brazilian e-Science‌ Workshop - SBCOPENLIBBreSci‌​‌ 2025 - XIX Brazilian​​ e-Science WorkshopAnais do​​​‌ XIX Brazilian e-Science Workshop‌ 2025Brasil, BrazilSeptember‌​‌ 2025, 1-8HAL​​DOI
• 49 inproceedingsG.​​​‌Giulio Martellucci, H.‌Hervé Goëau, P.‌​‌Pierre Bonnet, F.​​Fabrice Vinatier and A.​​​‌Alexis Joly. Overview‌ of PlantCLEF 2025: Multi-Species‌​‌ Plant Identification in Vegetation​​ Quadrat Images ⋆ Notebook​​​‌ for the LifeCLEF Lab‌ at CLEF 2025.‌​‌CLEF 2025 - Working​​ Notes of the Conference​​​‌ and Labs of the‌ Evaluation Forum4038CEUR‌​‌ - Workshop ProceedingsMadrid,​​ SpainCEUR2025,​​​‌ 2942-2954HALback to‌ text
• 50 inproceedingsR.‌​‌Rodrigo Parracho, F.​​Fernando Alexandrino, M.​​​‌Matheus de Souza Figueiredo‌, L.Lucas Pereira‌​‌ da Silva, B.​​Bruno Dutra de Macedo​​​‌, A.Arthur Lamblet‌ Vaz, D.Davi‌​‌ Louback, V. C.​​Victor Coculilo Desouzart,​​​‌ R.Rebecca Salles,‌ F.Fabio Porto,‌​‌ D.Diego Carvalho and​​ E.Eduardo Ogasawara.​​​‌ Leveraging Large Language Models‌ for Time Series Prediction‌​‌ on Low-Frequency Data.​​​‌SBBD 2025 - 40º​ Simpósio Brasileiro de Banco​‌ de DadosFortaleza, Brazil​​Sociedade Brasileira de Computação​​​‌ - SBC2025,​ 196-208HALDOI
• 51​‌ inproceedingsL.Lukáš Picek​​, C.César Leblanc​​​‌, T.Théo Larcher​, M.Maximilien Servajean​‌, P.Pierre Bonnet​​ and A.Alexis Joly​​​‌. Overview of GeoLifeCLEF​ 2025: Plant Species Presence​‌ Prediction with Environmental and​​ High-resolution Remote Sensing Data​​​‌.CLEF 2025 -​ Working Notes of the​‌ Conference and Labs of​​ the Evaluation Forum4038​​​‌CEUR Workshop ProceedingsMadrid,​ SpainCEUR2025,​‌ 2932-2941HALback to​​ text
• 52 inproceedingsE.​​​‌Edson Pinto Sobrinho,​ J.Jéssica Souza,​‌ J.Janio Lima,​​ L.Lucas Giusti,​​​‌ E.Eduardo Bezerra,​ R.Rafaelli Coutinho,​‌ L.Lais Baroni,​​ E.Esther Pacitti,​​​‌ F.Fabio Porto,​ K.Kele Belloze and​‌ E.Eduardo Ogasawara.​​ Fine-Tuning Detection Criteria for​​​‌ Enhancing Anomaly Detection in​ Time Series.SBBD​‌ 2025 - 40º Simpósio​​ Brasileiro de Banco de​​​‌ DadosFortaleza, BrazilSociedade​ Brasileira de Computação -​‌ SBC2025, 209-221​​HALDOIback to​​​‌ text
• 53 inproceedingsR.​Rebecca Salles, B.​‌Benoit Lange, R.​​Reza Akbarinia, F.​​​‌Florent Masseglia and E.​Esther Pacitti. Energy​‌ Efficient Time Series Anomaly​​ Detection.ICECET 2025​​​‌ - International Conference on​ Electrical, Computer and Energy​‌ TechnologiesParis, FranceJuly​​ 2025HALback to​​​‌ text

Conferences without proceedings​

• 54 inproceedingsM.Meije​‌ Gawinowski, L.Laurène​​ Perthame, F.Frédéric​​​‌ Rees, C.Céline​ Richard-Molard, C.Christophe​‌ Pradal, P.Pierre​​ Barbillon and A.Alexandra​​​‌ Jullien. Méta-modélisation des​ interactions plante-plante en 3D​‌ : application à l'association​​ colza-féverole.Journées scientifiques​​​‌ 2025 du PEPR Agroécologie​ et NumériqueDijon, France​‌January 2025HAL
• 55​​ inproceedingsG.Gabriela Moraes​​​‌, F.Fabio Porto​, F.Federico Ulliana​‌, J.-F.Jean-François Baget​​, M.Michel Leclère​​​‌, P.Pierre Bisquert​, B.Bernardo Gonçalves​‌ and P.Patrick Valduriez​​. Gypscie-KG: Building a​​​‌ Logic-Based Approach for Knowledge​ Graph Data Integration View​‌ in ML Systems.​​LAGO 2025 – LLMs,​​​‌ Análise de Grafos e​ Ontologias (SBBD)Fortaleza, Brazil​‌September 2025HALback​​ to text

Scientific books​​​‌

• 56 bookE. S.​Eduardo S. Ogasawara,​‌ R.Rebecca Salles,​​ F.Fabio Porto and​​​‌ E.Esther Pacitti.​ Event Detection in Time​‌ Series.Synthesis Lectures​​ on Data Management (SLDM)​​​‌SpringerFebruary 2025,​ 1-178HALback to​‌ text

Scientific book chapters​​

• 57 inbookL.Lukáš​​​‌ Picek, S.Stefan​ Kahl, H.Hervé​‌ Goëau, L.Lukáš​​ Adam, T.Théo​​​‌ Larcher, C.Cesar​ Leblanc, M.Maximilien​‌ Servajean, K.Klára​​ Janoušková, J.Jiří​​​‌ Matas, V.Vojtěch​ Čermák, K.Kostas​‌ Papafitsoros, R.Robert​​ Planqué, W.-P.Willem-Pier​​​‌ Vellinga, H.Holger​ Klinck, T.Tom​‌ Denton, J.Juan​​ Sebastián Cañas, G.​​​‌Giulio Martellucci, F.​Fabrice Vinatier, P.​‌Pierre Bonnet and A.​​Alexis Joly. Overview​​ of LifeCLEF 2025: Challenges​​​‌ on Species Presence Prediction‌ and Identification, and Individual‌​‌ Animal Identification.16089​​Experimental IR Meets Multilinguality,​​​‌ Multimodality, and Interaction: 16th‌ International Conference of the‌​‌ CLEF Association, CLEF 2025​​Lecture Notes in Computer​​​‌ ScienceSpringer Nature Switzerland‌January 2026, 338-362‌​‌HALDOIback to​​ text

Doctoral dissertations and​​​‌ habilitation theses

• 58 thesis‌M.Matteo Contini.‌​‌ Multi-scale mapping of changes​​ in tropical reefs.​​​‌Université de MontpellierNovember‌ 2025HALback to‌​‌ text

Reports & preprints​​

• 59 miscE.Eva​​​‌ Coindre, R.Romain‌ Boulord, L.Laurine‌​‌ Chir, V.Virgilio​​ Freitas, M.Maxime​​​‌ Ryckewaert, T.Thomas‌ Laisné, V.Virginie‌​‌ Bouckenooghe, M.Maëlle​​ Lis, L.Llorenç​​​‌ Cabrera-Bosquet, A.Agnès‌ Doligez, T.Thierry‌​‌ Simonneau, B.Benoît​​ Pallas, A.Aude​​​‌ Coupel-Ledru and V.Vincent‌ Segura. Robustness of‌​‌ high-throughput prediction of leaf​​ ecophysiological traits using near​​​‌ infrared spectroscopy and poro-fluorometry‌.February 2025HAL‌​‌DOI
• 60 reportG.​​Guillaume Coulaud, R.​​​‌Reza Akbarinia, A.‌Audrey Brouillet and F.‌​‌Florent Masseglia. Leveraging​​ Data Seasonality and Matrix​​​‌ Profile for Anomaly Detection:‌ Application to Climate Time‌​‌ Series.RR-9572Inria​​January 2025HALback​​​‌ to text
• 61 misc‌T.Tiffany Ding,‌​‌ J.-B.Jean-Baptiste Fermanian and​​ J.Joseph Salmon.​​​‌ Conformal Prediction for Long-Tailed‌ Classification.July 2025‌​‌HALback to text​​
• 62 miscK.Kian​​​‌ Faizi, P.Preyanka‌ Mehta, A.Amy‌​‌ Maida, T.Taylor​​ Humphreys, E.Elizabeth​​​‌ Berrigan, L.Leo‌ Mc Kee-Reid, R.‌​‌Robbie Mc Corkell,​​ A.Arush Tagade,​​​‌ J.Jessica Rumbelow,‌ J.Julia Showalter,‌​‌ L.Lukas Brent,​​ C.Clementine Coroenne,​​​‌ A.Audrey Rigaud,‌ A.Arjun Chandrasekhar,‌​‌ S.Saket Navlakha,​​ A.Antoine Martin,​​​‌ C.Christophe Pradal,‌ S.Sanghwa Lee,‌​‌ W.Wolfgang Busch and​​ M. P.Matthieu Pierre​​​‌ Platre. Growth Cost‌ and Transport Efficiency Tradeoffs‌​‌ Define Root System Optimization​​ Across Varying Developmental Stages​​​‌ and Environments in Arabidopsis‌.July 2025,‌​‌ 1-25HALDOI
• 63​​ miscH.Hanna Huitu​​​‌, T.-E.Tor-Einar Skog‌, C.Christophe Pradal‌​‌, A.Antonio Calatayud​​, T.Tor Skaslien​​​‌, B.Brita Linnestad‌, A.Ari Ronkainen‌​‌, C.Christian Fournier​​, M.Marc Labadie​​​‌, D.Dave Skirvin‌, M.Matti Pastell‌​‌, D.David Melchior​​, J. T.Johannes​​​‌ Tobiassen Langvatn and B.‌Berit Nordskog. Developing‌​‌ a generic DSS model​​ metadata catalogue and APIs​​​‌ for crop protection.‌August 2025HALDOI‌​‌back to text
• 64​​ miscM.Maxime Ryckewaert​​​‌, D.Diego Marcos‌, C.Christophe Botella‌​‌, M.Maximilien Servajean​​, P.Pierre Bonnet​​​‌ and A.Alexis Joly‌. Applying the maximum‌​‌ entropy principle to neural​​ networks enhances multi-species distribution​​​‌ models.January 2026‌HAL

Other scientific publications‌​‌

• 65 inproceedingsR.Romane​​ Dubois, L.Lydia​​​‌ Bousset, M.Melen‌ Leclerc, N.Nicolas‌​‌ Parisey and A.Alexis​​​‌ Joly. Weakly supervised​ segmentation of leaf symptoms​‌ in field conditions.​​Workshop Franco-Britannique organisé par​​​‌ le réseau « Modélisation​ et statistique pour la​‌ santé des animaux et​​ des plantes »Paris,​​​‌ FranceOctober 2025HAL​
• 66 inproceedingsT.Tristan​‌ Gérault, R.Romain​​ Barillot, C.Christophe​​​‌ Pradal, M.Marion​ Gauthier, C.Céline​‌ Richard-Molard, B.Bruno​​ Andrieu, A.Alexandra​​​‌ Jullien and F.Frédéric​ Rees. Modelling C​‌ and N root-soil exchanges​​ with 3D root and​​​‌ shoot growth based on​ plant ecophysiology: the Wheat-BRIDGES​‌ approach.Rhizosphere 6​​ - Rooting for Earth​​​‌Edinburgh, United Kingdom2025​HAL
• 67 miscI.​‌Ilyass Moummad. Audiocarnet​​ - Deep Representation Learning​​​‌ from Unlabeled Bioacoustic Data​.November 2025HAL​‌

Scientific popularization

• 68 inbook​​I.Ioana Manolescu and​​​‌ P.Patrick Valduriez.​ De nouvelles architectures pour​‌ les Big Data.​​Le calcul à découvert​​​‌CNRS EditionsJanuary 2025​HAL

Software

• 69 software​‌F.Frédéric Rees,​​ C.Christophe Pradal,​​​‌ M.Marion Gauthier and​ T.Tristan Gérault.​‌ RhizoDep: a Functional-Structural Root​​ Model to simulate rhizodeposition​​​‌.1.0.0April 2025​ lic: CeCILL-C.HAL​‌Software HeritageVCS