Artificial Intelligence and Modeling for Understanding Oceans and Climate Change
about
There is strong scientific evidence on the adverse effects of climate change on the global ocean. These changes will have a drastic impact on almost all life forms in the oceans with further consequences on food security, the ecosystems in coastal and inland communities. Despite these impacts, scientific data and infrastructures are still lacking to understand better and quantify the consequence of these perturbations on the marine ecosystem. It is necessary not only to gather more data but also to develop and apply state-of-the-art mechanisms capable of turning this data into effective knowledge, policies, and action. This is where artificial intelligence, machine learning, and modeling tools are called for.
This Inria ChallengeOcéanIA aims at developing new artificial intelligence and mathematical modeling tools to contribute to the understanding of the structure, functioning, underlying mechanisms, and dynamics of the oceans and their role in regulating and sustaining the biosphere, and tackling the climate change. OcéanIA is then an opportunity to structure Inria’s contributions around a global scientific challenge in the convergence of Artificial Intelligence, Biodiversity & Climate Change.
The goals of the project are structured in two directions. One that gathers the work from computer science and applied math to meet the challenges of the problem. The other focuses on applying the results of the first in multi-disciplinary application contexts.
The goals are discussed in depth in the project white book:
Sanchez-Pi, N., Martí, L., Abreu, A., Bernard, O., de Vargas, C., Eveillard, D., Maass, A., Marquet, P. A., Sainte-Marie, J., Salomon, J., Schoenauer, M., & Sebag, M. (2021). OcéanIA: AI, Data, and Models for Understanding the Ocean and Climate Change. (N. Sanchez-Pi & L. Martí, Eds.). Lille, Paris, Saclay, Santiago, Sophia-Antipolis: Inria – Institut national de recherche en sciences et technologies du numérique.download pdf
Computer science and math objectives
Data governance, curation, and availability
Data governance policy for marine biology and oceanographic data, consolidated access to data, and scientific computing software stacks.
Structured and graph-based neural networks
Model depth, models scalability, graph topological heterogeneity, and dynamic graphs.
Learning and adaptation in small data contexts
Transfer learning and domain adaptation, active and few-shot learning, and multi-source and multi-task learning deep neural models.
Causality and explainable models in AI
Causal inference, explainable AI and adversarial machine learning, interpretable shadow models and causal inference for understanding internal representations.
Model-driven and data-driven integration and hybrids
Learning PDEs from data, understanding neural network learning dynamics, and hybrid models combining PDE solvers and neural networks.
Development, calibration, and validation of mechanistic models
Identifiability issues, metabolic model reduction, and Navier-Stokes equations: From Eulerian to Lagrangian.
Multi-disciplinary applied objectives
Biodiversity and ecosystem functioning
Meta-metabolic modeling, phytoplankton biodiversity concerning temperature, present, and future, Data assimilation in biogeochemical models: Predicting the future.
Understanding plankton communities using AI, ML, and vision
Plankton identification from satellite images, connecting images and genomic features, anomaly detection, and explainable AI for automatic plankton discovery.
Interrelation of the objectives.
team
The OcéanIA team has diverse combination of skills, experience, and interests, something that is necessary to address a research-intensive and multi-disciplinary project such as this one.
Workshop AIMOCC 2024 at ECAI 2024 – AI: Modeling Oceans and Climate Change Workshop at ECAI 2024. Sánchez-Pi, N. and Martí, L. (eds). October 16-24, 2024. more information
Past meetings and workshops
OcéanIA 2023 annual meeting. February 23, 2024. more information
Workshop AIMOCC 2022 at IJCAI 2022 – AI: Modeling Oceans and Climate Change Workshop at IJCAI 2022. Sánchez-Pi, N. and Martí, L. (eds). July 23-29, 2022. online proceedings
OcéanIA Challenge at IJCAI 2022: AI methods for determining ocean ecosystems from space: Combining genomic information, microscopic and satellite imagery. Sánchez-Pi, N. and Martí, L. (eds). July 23-29, 2022. more information
Workshop AIMOCC 2021 at ICLR 2021 – AI: Modeling Oceans and Climate Change Workshop at ICLR 2021. Sánchez-Pi, N. and Martí, L. (eds). May 7, 2021. online proceedings
Software
Click here to see the software being developed within the project.
Publications
Journal Articles
Carrillo, H., de Wolff, T., Martí, L., & Sanchez-Pi, N. (2023). Evolutionary multi-objective physics-informed neural networks: The MOPINNs approach. AI Communications, 1–13.
doi: 10.3233/aic-230073bibtex
@article{10.3233/aic-230073,
title = {Evolutionary multi-objective physics-informed neural networks: {T}he {MOPINN}s approach},
issn = {0921-7126},
url = {http://dx.doi.org/10.3233/AIC-230073},
doi = {10.3233/aic-230073},
journal = {AI Communications},
publisher = {IOS Press},
author = {Carrillo, Hugo and de~Wolff, Taco and Mart\'{i}, Luis and Sanchez-Pi, Nayat},
year = {2023},
month = dec,
pages = {1--13}
}
Arroyo, J. I., Dı́ez Beatriz, Kempes, C. P., West, G. B., & Marquet, P. A. (2022). A general theory for temperature dependence in biology. Proceedings of the National Academy of Sciences of the United
States of America, 119(30), e2119872119.
doi: 10.1073/pnas.2119872119abstractbibtex
At present, there is no simple, first principles-based, and
general model for quantitatively describing the full range of
observed biological temperature responses. Here we derive a
general theory for temperature dependence in biology based on
Eyring-Evans-Polanyi’s theory for chemical reaction rates.
Assuming only that the conformational entropy of molecules
changes with temperature, we derive a theory for the
temperature dependence of enzyme reaction rates which takes
the form of an exponential function modified by a power law
and that describes the characteristic asymmetric curved
temperature response. Based on a few additional principles,
our model can be used to predict the temperature response
above the enzyme level, thus spanning quantum to classical
scales. Our theory provides an analytical description for the
shape of temperature response curves and demonstrates its
generality by showing the convergence of all temperature
dependence responses onto universal relationships-a universal
data collapse-under appropriate normalization and by
identifying a general optimal temperature, around 25 Celsius,
characterizing all temperature response curves. The model
provides a good fit to empirical data for a wide variety of
biological rates, times, and steady-state quantities, from
molecular to ecological scales and across multiple taxonomic
groups (from viruses to mammals). This theory provides a
simple framework to understand and predict the impact of
temperature on biological quantities based on the first
principles of thermodynamics, bridging quantum to classical
scales.
@article{Arroyo2022-fw,
title = {A general theory for temperature dependence in biology},
author = {Arroyo, Jos{\'e} Ignacio and D{\'\i}ez, Beatriz and Kempes, Christopher P and West, Geoffrey B and Marquet, Pablo A},
affiliation = {Departamento de Ecolog{\'\i}a, Facultad de Ciencias
Biol{\'o}gicas, Pontificia Universidad Cat{\'o}lica de Chile,
CP 8331150 Santiago, Chile. The Santa Fe Institute, Santa Fe,
NM 87501. Departamento de Gen{\'e}tica Molecular y
Microbiolog{\'\i}a, Facultad de Ciencias Biol{\'o}gicas,
Pontificia Universidad Cat{\'o}lica de Chile, CP 8331150
Santiago, Chile. Center for Climate and Resilience Research,
FONDAP (Fondo de Financiamiento de Centros de
Investigaci{\'o}n en {\'A}reas Prioritarias), University of
Chile, CP 8370449 Santiago, Chile. Center for Genome
Regulation, FONDAP, Faculty of Science, University of Chile,
CP 7800003 Santiago, Chile. Instituto de Ecolog{\'\i}a y
Biodiversidad, CP 7800003 Santiago, Chile. Centro de Cambio
Global Universidad Cat{\'o}lica, Facultad de Ciencias
Biol{\'o}gicas, Pontificia Universidad Cat{\'o}lica de Chile,
CP 8331150 Santiago, Chile. Instituto de Sistemas Complejos de
Valpara{\'\i}so, CP 2340000 Valpara{\'\i}so, Chile. Centro de
Modelamiento Matem{\'a}tico, Universidad de Chile,
International Research Laboratory 2807, CNRS, CP 8370456
Santiago, Chile.},
journal = {Proceedings of the National Academy of Sciences of the United
States of America},
volume = {119},
number = {30},
pages = {e2119872119},
month = jul,
year = {2022},
url = {http://dx.doi.org/10.1073/pnas.2119872119},
keywords = {metabolic theory; scaling; temperature kinetics},
language = {en},
issn = {0027-8424, 1091-6490},
pmid = {35858416},
doi = {10.1073/pnas.2119872119},
pmc = {PMC9335213}
}
Aguayo, J., & Lincopi, H. C. (2022). Analysis of Obstacles Immersed in Viscous Fluids Using Brinkman’s Law for Steady Stokes and Navier-Stokes Equations. SIAM Journal Applied Mathematics, 82(4), 1369–1386.
doi: 10.1137/20M138569Xabstractbibtex
From the steady Stokes and Navier-Stokes models, a penalization method has been considered by several authors for approximating those fluid equations around obstacles. In this work, we present a justification for using fictitious domains to study obstacles immersed in incompressible viscous fluids through a simplified version of Brinkman’s law for porous media. If the scalar function psi is considered as the inverse of permeability, it is possible to study the singularities of ψas approximations of obstacles (when ψtends to ∞) or of the domain corresponding to the fluid (when ψ= 0 or is very close to 0). The strong convergence of the solution of the perturbed problem to the solution of the strong problem is studied, also considering error estimates that depend on the penalty parameter, for fluids modeled with both the Stokes and Navier–Stokes equations with inhomogeneous boundary conditions. A numerical experiment is presented that validates this result and allows us to study the application of this perturbed problem simulation of flows and the identification of obstacles.
@article{10.1137/20M138569X,
author = {Aguayo, Jorge and Lincopi, Hugo Carrillo},
title = {Analysis of Obstacles Immersed in Viscous Fluids Using Brinkman's Law for Steady Stokes and Navier-Stokes Equations},
year = {2022},
issue_date = {January 2022},
publisher = {Society for Industrial and Applied Mathematics},
address = {USA},
volume = {82},
number = {4},
issn = {0036-1399},
url = {https://doi.org/10.1137/20M138569X},
doi = {10.1137/20M138569X},
journal = {SIAM Journal Applied Mathematics},
month = jan,
pages = {1369--1386},
numpages = {18},
keywords = {Stokes problem, Navier-Stokes, obstacles, penalization, Brinkman's law, 35, 65, 76}
}
Lira, H., Martí, L., & Sanchez-Pi, N. (2022). A Graph Neural Network with Spatio-Temporal Attention for Multi-Sources Time Series Data: An Application to Frost Forecast. Sensors, 22(4).
doi: 10.3390/s22041486
hal: hal-03541565abstractbibtex
Frost forecast is an important issue in climate research because of its economic impact on several industries. In this study, we propose GRAST-Frost, a graph neural network (GNN) with spatio-temporal architecture, which is used to predict minimum temperatures and the incidence of frost. We developed an IoT platform capable of acquiring weather data from an experimental site, and in addition, data were collected from 10 weather stations in close proximity to the aforementioned site. The model considers spatial and temporal relations while processing multiple time series simultaneously. Performing predictions of 6, 12, 24, and 48 h in advance, this model outperforms classical time series forecasting methods, including linear and nonlinear machine learning methods, simple deep learning architectures, and nongraph deep learning models. In addition, we show that our model significantly improves on the current state of the art of frost forecasting methods.
@article{lira-2022:graph-nn-frost,
title = {A Graph Neural Network with Spatio-Temporal Attention for Multi-Sources Time Series Data: {A}n Application to Frost Forecast},
author = {Lira, Hernan and Mart\'{i}, Luis and Sanchez-Pi, Nayat},
year = {2022},
journal = {Sensors},
volume = {22},
number = {4},
doi = {10.3390/s22041486},
issn = {1424-8220},
url = {https://www.mdpi.com/1424-8220/22/4/1486},
article-number = {1486},
hal_id = {hal-03541565},
hal_version = {v1}
}
Demory, D., Weitz, J. S., Baudoux, A.-C., Touzeau, S., Simon, N., Rabouille, S., Sciandra, A., & Bernard, O. (2021). A thermal trade-off between viral production and degradation drives virus-phytoplankton population dynamics. Ecology Letters, 24(6), 1133–1144.
doi: 10.1111/ele.13722bibtex
@article{demory-2021--thermal-trade-off,
author = {Demory, David and Weitz, Joshua S. and Baudoux, Anne-Claire and Touzeau, Suzanne and Simon, Natalie and Rabouille, Sophie and Sciandra, Antoine and Bernard, Olivier},
doi = {10.1111/ele.13722},
editor = {Lafferty, Kevin},
journal = {Ecology Letters},
month = apr,
number = {6},
pages = {1133--1144},
publisher = {Wiley},
title = {A thermal trade-off between viral production and degradation drives virus-phytoplankton population dynamics},
url = {https://doi.org/10.1111/ele.13722},
volume = {24},
year = {2021}
}
Books
The OcénIA Project White Book: Sanchez-Pi, N., Martí, L., Abreu, A., Bernard, O., de Vargas, C., Eveillard, D., Maass, A., Marquet, P. A., Sainte-Marie, J., Salomon, J., Schoenauer, M., & Sebag, M. (2021). OcéanIA: AI, Data, and Models for Understanding the Ocean and Climate Change. (N. Sanchez-Pi & L. Martí, Eds.). Lille, Paris, Saclay, Santiago, Sophia-Antipolis: Inria – Institut national de recherche en sciences et technologies du numérique.
hal: hal-03274323pdfbibtex
@book{oceania-white-book-2021,
title = {{Oc\'{e}anIA}: {AI}, Data, and Models for Understanding the Ocean and Climate Change},
author = {Sanchez-Pi, Nayat and Mart\'{i}, Luis and Abreu, Andr\'{e} and Bernard, Olivier and de~Vargas, Colomban and Eveillard, Damien and Maass, Alejandro and Marquet, Pablo~A. and Sainte-Marie, Jacques and Salomon, Julien and Schoenauer, Marc and Sebag, Mich\`{e}le},
year = {2021},
month = jul,
publisher = {Inria -- Institut national de recherche en sciences et technologies du num\'{e}rique},
address = {Lille, Paris, Saclay, Santiago, Sophia-Antipolis},
url = {http://oceania.inria.cl},
note = {The Oc\'{e}nIA Project White Book},
editor = {Sanchez-Pi, Nayat and Mart\'{i}, Luis},
eprint = {hal-01882235},
eprinttype = {hal},
hal_id = {hal-03274323},
hal_version = {v1}
}
Conference papers
Vasconcellos, E. C., Sampaio, R. M., Araújo, A. P. D., Gonzales Clua, E. W., Preux, P., Guerra, R., Gonçalves, L. M. G., Martí, L., Lira, H., & Sanchez-Pi, N. (2023). Reinforcement-learning robotic sailboats: Simulator and preliminary results. In NeurIPS 2023 Workshop on Robot Learning Workshop: Pretraining, Fine-Tuning, and Generalization with Large Scale Models. New Orleans, United States.
hal: hal-04395990pdfbibtex
@inproceedings{vasconcellos:hal-04395990,
title = {Reinforcement-learning robotic sailboats: {S}imulator and preliminary results},
author = {Vasconcellos, Eduardo Charles and Sampaio, Ronald M and Ara{\'u}jo, Andr{\'e} P D and Gonzales Clua, Esteban Walter and Preux, Philippe and Guerra, Raphael and Gon{\c c}alves, Luiz M G and Mart{\'i}, Luis and Lira, Hernan and Sanchez-Pi, Nayat},
url = {https://inria.hal.science/hal-04395990},
booktitle = {{NeurIPS 2023 Workshop on Robot Learning Workshop: Pretraining, Fine-Tuning, and Generalization with Large Scale Models}},
address = {New Orleans, United States},
year = {2023},
month = dec,
keywords = {artificial intelligence, reinforcement learning, unmanned surface vessel},
pdf = {https://inria.hal.science/hal-04395990/file/neurips_wrl2023.pdf},
hal_id = {hal-04395990},
hal_version = {v1}
}
Ferreira de Moraes, R., Evangelista, R. dos S., Pereira, A. L. da S., Toledo, Y. P., Fernandes, L. A. F., & Martí, L. (2023). Heuristics to reduce linear combinations of activation functions to improve image classification. In 2023 36th SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI) (pp. 169–174). IEEE.
doi: 10.1109/SIBGRAPI59091.2023.10347043
hal: hal-04396390abstractbibtex
Image classification is one of the classical problems in computer vision, and CNNs (Convolutional Neural Networks) are widely used for this task. However, the choice of a CNN can vary depending on the chosen dataset. In this context, we have trainable activation functions that are crucial in CNNs and adapt to the data. One technique for constructing these functions is to write them as a linear combination of other activation functions, where the coefficients of this combination are learned during training. However, if we have a large number of activation functions to combine, the computational cost can be very high, and manually testing and choosing these functions may be impractical, depending on the number of available activation functions. To alleviate the difficulty of choosing which activation functions should be part of the linear combination, we propose two heuristics: Linear Combination Approximator by Coefficients (LCAC) and Major and Uniform Coefficient Extractor (MUCE). Our heuristics provide an efficient selection of a subset of activation functions so that their results are better or equivalent to the linear combination that uses all 34 available activation functions in our experiments (C34), considering the image classification problem. Compared to the C34 function, the LCAC function was better or equivalent in 62.5%, and the MUCE function in 87.5% of the conducted experiments.
@inproceedings{10347043,
author = {Ferreira de Moraes, Rog\'{e}rio and Evangelista, Raphael dos S. and Pereira, Andre Luiz da S. and Toledo, Yanexis Pupo and Fernandes, Leandro A. F. and Mart\'{i}, Luis},
booktitle = {2023 36th SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI)},
title = {Heuristics to reduce linear combinations of activation functions to improve image classification},
year = {2023},
volume = {},
number = {},
pages = {169--174},
doi = {10.1109/SIBGRAPI59091.2023.10347043},
issn = {2377-5416},
location = {Rio Grande, Brazil},
publisher = {IEEE},
keywords = {Training ; Graphics ; Computer vision ; Computational efficiency ; Convolutional neural networks ; Task analysis ; Image classification},
hal_id = {hal-04396390},
hal_version = {v1},
month = nov
}
Fierro Ulloa, J. I., & Bernard, O. (2023). Physic-informed neural networks for microalgae modeling. In ECCE 14 & ECAB 7 - 14th European Congress of Chemical Engineering and 7th European Congress of Applied Biotechnology. Berlin, Germany.
hal: hal-04390804bibtex
@inproceedings{bernard2023physic,
title = {Physic-informed neural networks for microalgae modeling},
author = {Fierro Ulloa, Joel Ignacio and Bernard, Olivier},
url = {https://inria.hal.science/hal-04390804},
booktitle = {{ECCE} 14 \& {ECAB} 7 - 14th European Congress of Chemical Engineering and 7th European Congress of Applied Biotechnology},
address = {Berlin, Germany},
year = {2023},
month = sep,
hal_id = {hal-04390804},
hal_version = {v1}
}
de Wolff, T., Carrillo Lincopi, H., Sanchez-Pi, N., & Martí, L. (2022). MOPINNs: An Evolutionary Multi-Objective Approach to Physics-Informed Neural Networks. In Proceedings of The Genetic and Evolutionary Computation Conference 2022 (GECCO ’22). New York, NY, USA: Association for Computing Machinery.
doi: 10.1145/3520304.3529071bibtex
@inproceedings{10.1145/3520304.3529071,
title = {{MOPINNs}: {A}n Evolutionary Multi-Objective Approach to Physics-Informed Neural Networks},
author = {de Wolff, Taco and Carrillo Lincopi, Hugo and Sanchez-Pi, Nayat and Mart\'{i}, Luis},
year = {2022},
month = jul,
booktitle = {Proceedings of The Genetic and Evolutionary Computation Conference 2022 (GECCO '22)},
location = {Boston, MA},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
series = {GECCO'22},
doi = {10.1145/3520304.3529071},
isbn = {781450392686},
url = {https://doi.org/10.1145/3520304.3529071},
numpages = {4}
}
Carrillo, H., de Wolff, T., Martí, L., & Sanchez-Pi, N. (2022). Towards Optimally Weighted Physics-Informed Neural Networks in Ocean Modelling. In N. Sanchez-Pi & L. Martí (Eds.), Workshop AI: Modeling Oceans and Climate Change (AIMOCC 2022) of the 31st International Joint Conference on Artificial Intelligence and the 25th European Conference on Artificial Intelligence (IJCAI-ECAI 2022).bibtex
@inproceedings{hugo-vienna,
author = {Carrillo, Hugo and de Wolff, Taco and Mart\'{i}, Luis and Sanchez-Pi, Nayat},
booktitle = {Workshop AI: Modeling Oceans and Climate Change (AIMOCC 2022) of the 31st International Joint Conference on Artificial Intelligence and the 25th European Conference on Artificial Intelligence (IJCAI-ECAI 2022)},
title = {Towards Optimally Weighted Physics-Informed Neural Networks in Ocean Modelling},
location = {Vienna, Austria},
year = {2022},
month = jul,
editor = {Sanchez-Pi, Nayat and Mart\'{i}, Luis}
}
de Wolff, T., Carrillo, H., Martí, L., & Sanchez-Pi, N. (2022). Optimal Architecture Discovery for Physics-Informed Neural Networks. In A. C. Bicharra Garcia, M. Ferro, & J. C. Rodríguez Ribón (Eds.), Advances in Artificial Intelligence — IBERAMIA 2022 (pp. 77–88). Cham: Springer International Publishing.
doi: 10.1007/978-3-031-22419-5_7abstractbibtex
Physics-informed neural networks allow the neural network to be trained by both the training data and prior domain knowledge about the physical system that models the data. In particular, it has a loss function for the data and the physics, where the latter is the deviation from a partial differential equation describing the system. Conventionally, both loss functions are combined by a weighted sum, but this leaves the optimal weight unknown. Additionally, it is necessary to find the optimal architecture of the neural network. In our work, we propose a multi-objective optimization approach to find the optimal value for the loss function weighting, as well as the optimal activation function, number of layers, and number of neurons for each layer. We validate our results on the Burgers and wave equations and show that we are able to find accurate approximations of the solution using optimal hyperparameters.
@inproceedings{10.1007/978-3-031-22419-5_7,
address = {Cham},
author = {de Wolff, Taco and Carrillo, Hugo and Mart\'{i}, Luis and Sanchez-Pi, Nayat},
booktitle = {Advances in Artificial Intelligence --- {IBERAMIA} 2022},
editor = {Bicharra Garcia, Ana Cristina and Ferro, Mariza and Rodr\'{i}guez Rib\'{o}n, Julio Cesar},
isbn = {978-3-031-22419-5},
pages = {77--88},
publisher = {Springer International Publishing},
title = {Optimal Architecture Discovery for Physics-Informed Neural Networks},
year = {2022},
doi = {10.1007/978-3-031-22419-5_7}
}
Ferreira de Moraes, R., Evangelista, R. dos S., Fernandes, L. A. F., & Martí, L. (2021). GCOOD: A Generic Coupled Out-of-Distribution Detector for Robust Classification. In 2021 34th SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI) (pp. 409–416).
doi: 10.1109/SIBGRAPI54419.2021.00062
hal: hal-04396403abstractbibtex
Neural networks have achieved high degrees of accuracy in classification tasks. However, when an out-of-distribution (OOD) sample (i.e., entries from unknown classes) is submitted to the classification process, the result is the association of the sample to one or more of the trained classes with different degrees of confidence. If any of these confidence values are more significant than the user-defined threshold, the network will mislabel the sample, affecting the model credibility. The definition of the acceptance threshold itself is a sensitive issue in the face of the classifier’s overconfidence. This paper presents the Generic Coupled OOD Detector (GCOOD), a novel Convolutional Neural Network (CNN) tailored to detect whether an entry submitted to a trained classification model is an OOD sample for that model. From the analysis of the Softmax output of any classifier, our approach can indicate whether the resulting classification should be considered or not as a sample of some of the trained classes. To train our CNN, we had to develop a novel training strategy based on Voronoi diagrams of the location of representative entries in the latent space of the classification model and graph coloring. We evaluated our approach using ResNet, VGG, DenseNet, and SqueezeNet classifiers with images from the CIFAR-10 dataset.
@inproceedings{9643095,
author = {Ferreira de Moraes, Rog\'{e}rio and Evangelista, Raphael dos S. and Fernandes, Leandro A. F. and Mart\'{i}, Luis},
booktitle = {2021 34th SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI)},
title = {{GCOOD}: {A} Generic Coupled Out-of-Distribution Detector for Robust Classification},
year = {2021},
volume = {},
number = {},
pages = {409--416},
doi = {10.1109/SIBGRAPI54419.2021.00062},
issn = {2377-5416},
location = {Gramado, Rio Grande do Sul, Brazil},
keywords = {Training ; Graphics ; Neural networks ; Detectors ; Convolutional neural networks ; Task analysis ; Faces},
hal_id = {hal-04396403},
hal_version = {v1},
month = oct
}
Lira, H., Martí, L., & Sanchez-Pi, N. (2021). Frost forecasting model using graph neural networks with spatio-temporal attention. In N. Sanchez-Pi & L. Martí (Eds.), AI: Modeling Oceans and Climate Change Workshop at ICLR 2021. Santiago, Chile.
hal: hal-03259658pdfbibtex
@inproceedings{lira:hal-03259658,
title = {Frost forecasting model using graph neural networks with spatio-temporal attention},
author = {Lira, Hernan and Mart\'{i}, Luis and Sanchez-Pi, Nayat},
year = {2021},
month = may,
booktitle = {AI: Modeling Oceans and Climate Change Workshop at ICLR 2021},
address = {Santiago, Chile},
url = {https://hal.inria.fr/hal-03259658},
editor = {Sanchez-Pi, Nayat and Mart\'{i}, Luis},
hal_id = {hal-03259658},
hal_version = {v1},
pdf = {https://hal.inria.fr/hal-03259658/file/iclr2021_conference.pdf}
}
de Wolff, T., Carrillo, H., Martí, L., & Sanchez-Pi, N. (2021). Assessing Physics Informed Neural Networks in Ocean Modelling and Climate Change Applications. In N. Sanchez-Pi & L. Martí (Eds.), AI: Modeling Oceans and Climate Change Workshop at ICLR 2021. Santiago (Virtual), Chile.
hal: hal-03262684pdfbibtex
@inproceedings{dewolff:hal-03262684,
title = {Assessing Physics Informed Neural Networks in Ocean Modelling and Climate Change Applications},
author = {de Wolff, Taco and Carrillo, Hugo and Mart\'{i}, Luis and Sanchez-Pi, Nayat},
year = {2021},
month = may,
booktitle = {{AI}: {M}odeling Oceans and Climate Change Workshop at {ICLR} 2021},
address = {Santiago (Virtual), Chile},
url = {https://hal.inria.fr/hal-03262684},
editor = {Sanchez-Pi, Nayat and Mart\'{i}, Luis},
hal_id = {hal-03262684},
hal_version = {v1},
pdf = {https://hal.inria.fr/hal-03262684/file/iclr2021_conference.pdf}
}
Muñoz, A., Martí, L., & Sanchez-Pi, N. (2021). Data Governance, a Knowledge Model Through Ontologies. In R. Valencia-García, M. Bucaram-Leverone, J. Del Cioppo-Morstadt, N. Vera-Lucio, & E. Jácome-Murillo (Eds.), Technologies and Innovation (pp. 18–32). Cham: Springer International Publishing.
doi: 10.1007/978-3-030-88262-4_2abstractbibtex
Ontologies have emerged as a powerful tool for sharing knowledge, due to their ability to integrate them. A key challenge is the interoperability of data sources that do not have a common schema and that were collected, processed and analyzed under different methodologies. Data governance defines policies, organization and standards. Data governance focused on integration processes helps to define what is integrated, who does it and how it is integrated. The representation of this integration process implies that not only the elements involved in the integration of metadata and their data sets need to be represented, but also elements of coordination between people and knowledge domains need to be included. This paper shows the ontology that describes the data governance processes, the elements that make it up and their relationships. For its development, the methodology based on competency questions and definition of terms is used. The data governance ontology creates a context to support the interaction of different data sources. The ontology is instantiated by means of a case study for Data Governance in Mining Inspection for the Geology and Mining Service of the Chilean government.
@inproceedings{10.1007/978-3-030-88262-4_2,
title = {Data Governance, a Knowledge Model Through Ontologies},
author = {Mu\~{n}oz, Ana and Mart\'{i}, Luis and Sanchez-Pi, Nayat},
year = {2021},
booktitle = {Technologies and Innovation},
publisher = {Springer International Publishing},
address = {Cham},
pages = {18--32},
doi = {10.1007/978-3-030-88262-4_2},
isbn = {978-3-030-88262-4},
issn = {1865-0929},
editor = {Valencia-Garc\'{i}a, Rafael and Bucaram-Leverone, Martha and Del Cioppo-Morstadt, Javier and Vera-Lucio, N\'{e}stor and J\'{a}come-Murillo, Emma}
}
Sanchez-Pi, N., Martí, L., Abreu, A., Bernard, O., de Vargas, C., Eveillard, D., Maass, A., Marquet, P. A., Sainte-Marie, J., Salomon, J., Schoenauer, M., & Sebag, M. (2020). Artificial Intelligence, Machine Learning and Modeling for Understanding the Oceans and Climate Change. In D. Dao, E. Sherwin, P. Donti, L. Kaack, L. Kuntz, Y. Yusuf, D. Rolnick, C. Nakalembe, C. Monteleoni, & Y. Bengio (Eds.), Tackling Climate Change with Machine Learning workshop at NeurIPS 2020.
hal: hal-03138712pdfslidesabstractbibtex
The ongoing transformation of climate and biodiversity will have a drastic impact on almost all forms of life in the ocean with further consequences on food security, ecosystem services in coastal and inland communities. Despite these impacts, scientific data and infrastructures are still lacking to understand and quantify the consequences of these perturbations on the marine ecosystem. Understanding this phenomenon is not only an urgent but also a scientifically demanding task. Consequently, it is a problem that must be addressed with a scientific cohort approach, where multi-disciplinary teams collaborate to bring the best of different scientific areas. In this proposal paper, we describe our newly launched four-years project focused on developing new artificial intelligence, machine learning, and mathematical modeling tools to contribute to the understanding of the structure, functioning, and underlying mechanisms and dynamics of the global ocean symbiome and its relation with climate change. These actions should enable the understanding of our oceans and predict and mitigate the consequences of climate and biodiversity changes.
@inproceedings{sanchez-pi-et-al-2020--neurips,
title = {Artificial Intelligence, Machine Learning and Modeling for Understanding the Oceans and Climate Change},
author = {Sanchez-Pi, Nayat and Mart\'{i}, Luis and Abreu, Andr\'{e} and Bernard, Olivier and de~Vargas, Colomban and Eveillard, Damien and Maass, Alejandro and Marquet, Pablo~A. and Sainte-Marie, Jacques and Salomon, Julien and Schoenauer, Marc and Sebag, Mich\`{e}le},
year = {2020},
month = dec,
booktitle = {Tackling Climate Change with Machine Learning workshop at NeurIPS 2020},
url = {https://www.climatechange.ai/papers/neurips2020/93},
editor = {Dao, David and Sherwin, Evan and Donti, Priya and Kaack, Lynn and Kuntz, Lauren and Yusuf, Yumna and Rolnick, David and Nakalembe, Catherine and Monteleoni, Claire and Bengio, Yoshua},
hal_id = {hal-03138712},
hal_version = {v1},
pdf = {https://hal.archives-ouvertes.fr/hal-03138712/file/paper.pdf},
slides_pdf_url = {https://www.climatechange.ai/papers/neurips2020/93/slides.pdf}
}
Keynotes and talks
Fierro Ulloa, J. I. (2023). NeuralODEs for phytoplankton modeling. In Journées scientifiques Inria Chile 2023.bibtex
@conference{fierro-2023--jsic,
author = {Fierro Ulloa, Joel Ignacio},
booktitle = {Journ\'{e}es scientifiques Inria Chile 2023},
month = dec,
title = {{NeuralODEs} for phytoplankton modeling},
year = {2023},
location = {Santiago/Valpara\'{i}so/Concepci\'{o}n, Chile}
}
Keynote: Martí, L. (2023). Explainable AI for understanding plankton communities. In Journées scientifiques Inria Chile 2023.bibtex
Keynote: Martí, L. (2023). AI for (climate) good. In Escuela de Verano en Inteligencia Computacional. Facultad de Ciencias Físicas y Matemáticas, Universidad de Chile.bibtex
@conference{lmarti-2023--evic,
author = {Mart\'{i}, Luis},
booktitle = {Escuela de Verano en Inteligencia Computacional},
month = dec,
title = {{AI} for (climate) good},
year = {2023},
institution = {Facultad de Ciencias Físicas y Matemáticas, Universidad de Chile},
location = {Santiago de Chile, Chile},
note = {Keynote}
}
Valenzuela, L. (2023). Modeling global plankton communities via multinomics and ML approaches. In Journées scientifiques Inria Chile 2023.bibtex
@conference{valenzuela-2023--jsic,
author = {Valenzuela, Luis},
booktitle = {Journ\'{e}es scientifiques Inria Chile 2023},
month = dec,
title = {Modeling global plankton communities via multinomics and ML approaches},
year = {2023},
location = {Santiago/Valpara\'{i}so/Concepci\'{o}n, Chile}
}
Sanchez-Pi, N., & Martí, L. (2023). Opportunities in Chile: from the origin of the universe to the depths of the ocean. In Journées scientifiques Inria Chile 2023.bibtex
@conference{sanchez-pi-2023--jjss-inria,
author = {Sanchez-Pi, Nayat and Mart\'{i}, Luis},
booktitle = {Journ\'{e}es scientifiques Inria Chile 2023},
month = sep,
title = {Opportunities in {C}hile: from the origin of the universe to the depths of the ocean},
year = {2023},
location = {Inria Bordeaux, France},
url = {https://mediatheque.inria.fr/Mediatheque/embed/public/79139}
}
Carrillo, H. (2023). A Reduced Basis (RB) approach combined with Artificial Neural Networks (ANN) for optimal location of measurements for inverse problems in PDEs. In numhyp23: Numerical Methods for Hyperbolic Problems.bibtex
@conference{carrillo-2023--fr,
author = {Carrillo, Hugo},
booktitle = {numhyp23: {N}umerical Methods for Hyperbolic Problems},
month = jun,
title = {A Reduced Basis ({RB}) approach combined with Artificial Neural Networks ({ANN}) for optimal location of measurements for inverse problems in PDEs},
year = {2023},
url = {https://numhyp23.sciencesconf.org/},
location = {Bordeaux, France}
}
Keynote: Sanchez-Pi, N. (2023). AI, the Ocean and Climate Change. In KHIPU: Latin American Meeting on Artificial Intelligence.bibtex
@conference{sanchez-pi-2023--khipu,
author = {Sanchez-Pi, Nayat},
booktitle = {KHIPU: Latin American Meeting on Artificial Intelligence},
month = mar,
title = {{AI}, the Ocean and Climate Change},
year = {2023},
location = {Universidad de La Rep\'{u}blica, Uruguay},
url = {https://khipu.ai/khipu2023/},
note = {Keynote}
}
Invited tutorial: Sanchez-Pi, N., & Martí, L. (2021). Towards a Green AI: Evolutionary Solutions for an Ecologically Viable Artificial Intelligence. In Proceedings of the Genetic and Evolutionary Computation Conference Companion (pp. 1135–1140). New York, NY, USA: Association for Computing Machinery.
doi: 10.1145/3449726.3461428bibtex
@conference{10.1145/3449726.3461428,
title = {Towards a Green {AI}: {E}volutionary Solutions for an Ecologically Viable Artificial Intelligence},
author = {Sanchez-Pi, Nayat and Mart\'{i}, Luis},
year = {2021},
month = jul,
booktitle = {Proceedings of the Genetic and Evolutionary Computation Conference Companion},
location = {Lille, France},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
series = {GECCO'21},
pages = {1135--1140},
doi = {10.1145/3449726.3461428},
isbn = {9781450383516},
url = {https://doi.org/10.1145/3449726.3461428},
note = {Invited tutorial},
numpages = {6}
}
Keynote: Sanchez-Pi, N. (2020). OcéanIA: AI, Oceans and Climate Change. In A. Ruiz-Garcia, I. Arraut, J. M. Banda, I. Lopez-Francos, F. Latorre, P. Magalhães Braga, K. Caballero Barajas, S. H. Garrido Mejia, E. U. Moya-Sánchez, V. Fernandes Caridá, C. Miranda, G. Bejarano, & J. Ortega Caro (Eds.), LatinX in AI (LXAI) workshop at NeurIPS 2020.slidesview onlinebibtex
@conference{sanchez-pi-2020--latinxs-neurips,
title = {{Oc\'{e}anIA}: {AI}, Oceans and Climate Change},
author = {Sanchez-Pi, Nayat},
year = {2020},
month = dec,
booktitle = {LatinX in AI (LXAI) workshop at NeurIPS 2020},
url = {https://www.latinxinai.org/neurips-2020-about},
note = {Keynote},
editor = {Ruiz-Garcia, Ariel and Arraut, Ivan and Banda, Juan~M. and Lopez-Francos, Ignacio and Latorre, Fabian and Magalh\~{a}es Braga, Pedro and Caballero Barajas, Karla and Garrido Mejia, Sergio Hernan and Moya-S\'{a}nchez, Eduardo Ulises and Fernandes Carid\'{a}, Vinicius and Miranda, Carlos and Bejarano, Gissella and Ortega Caro, Josu\'{e}},
online_presentation_url = {https://slideslive.com/38942447/ai-oceans-and-climate-change}
}
Theses
Migus, L. (2023, December). Deep neural networks and partial differential equations (phdthesis). Sorbonne Université.
hal: tel-04336969pdfbibtex
@thesis{migus:tel-04336969,
title = {Deep neural networks and partial differential equations},
author = {Migus, L{\'e}on},
url = {https://theses.hal.science/tel-04336969},
number = {2023SORUS356},
institution = {Sorbonne Universit{\'e}},
supervisor = {Julien Salomnon and Patrick Gallinari},
year = {2023},
month = dec,
keywords = {Deep neural networks ; Differential equations ; Numerical schemes ; Machine learning ; R{\'e}seaux de neurones ; Equations diff{\'e}rentielles ; Sch{\'e}mas num{\'e}riques ; Apprentissage machine},
type = {phdthesis},
pdf = {https://theses.hal.science/tel-04336969/file/MIGUS_Leon_these_2023.pdf},
hal_id = {tel-04336969},
hal_version = {v1}
}
Ferreira de Moraes, R. (2023). Novas Abordagens para Função de Perda e Função de Ativação para Classificação de Imagens (phdthesis). Universidade Federal Fluminense.bibtex
@thesis{rogerio-phd,
author = {Ferreira de Moraes, Rog\'{e}rio},
title = {Novas Abordagens para Função de Perda e Função de Ativação para Classificação de Imagens},
type = {phdthesis},
institution = {Universidade Federal Fluminense},
year = {2023},
location = {Niterói, Brazil},
supervisor = {Leandro Fernandes and Luis Mart\'{i}}
}
Other activities
Martí, L. (2023, October). AI for (climate) good. Seminar Series of the Institute of Astrophysics Studies, Universidad Diego Portales. Institute of Astrophysics Studies, Universidad Diego Portales.bibtex
@misc{lmarti-2023--udp,
author = {Mart\'{i}, Luis},
booktitle = {Seminar Series of the Institute of Astrophysics Studies, Universidad Diego Portales},
month = oct,
title = {{AI} for (climate) good},
year = {2023},
institution = {Institute of Astrophysics Studies, Universidad Diego Portales},
location = {Santiago de Chile, Chile}
}
Valenzuela, L. (2023, September). Inteligencia Artificial y Genómica para comprender los océanos y el cambio climático. F\hatete de la Science 2023 Chile.bibtex
@misc{lvalenzuela-2023--fete-de-la-science,
author = {Valenzuela, Luis},
year = {2023},
month = sep,
title = {Inteligencia Artificial y Gen\'{o}mica para comprender los oc\'{e}anos y el cambio clim\'{a}tico},
booktitle = {F\hat{e}te de la Science 2023 Chile}
}
Project presentation to students: Valenzuela, L. (2023). OcéanIA. Santiago, Chile: Universidad Austral de Chile.bibtex
@misc{lvalenzuela-2023--uach,
author = {Valenzuela, Luis},
year = {2023},
title = {Oc\'{e}anIA},
note = {Project presentation to students},
address = {Santiago, Chile},
institution = {Universidad Austral de Chile}
}
proceedings
Sanchez-Pi, N., & Martí, L. (Eds.). (2022). AI: Modeling Oceans and Climate Change Workshop (AIMOCC 2022). Vienna, Austria: 31st International Joint Conference on Artificial Intelligence and the 25th European Conference on Artificial Intelligence (IJCAI-ECAI 2022).bibtex
@proceedings{sanchez-marti-2022:aimocc,
title = {{AI: Modeling Oceans and Climate Change Workshop (AIMOCC 2022)}},
year = {2022},
month = jul,
publisher = {31st International Joint Conference on Artificial Intelligence and the 25th European Conference on Artificial Intelligence (IJCAI-ECAI 2022)},
address = {Vienna, Austria},
url = {https://oceania.inria.cl/#aimocc-2022},
editor = {Sanchez-Pi, Nayat and Mart\'{i}, Luis}
}
Sanchez-Pi, N., & Martí, L. (Eds.). (2021). AI: Modeling Oceans and Climate Change Workshop (AIMOCC 2021). Santiago de Chile (Virtual): Tenth International Conference on Learning Representations (ICLR 2021).bibtex
@proceedings{sanchez-marti-2021:aimocc,
title = {{AI: Modeling Oceans and Climate Change Workshop (AIMOCC 2021)}},
year = {2021},
month = may,
publisher = {Tenth International Conference on Learning Representations (ICLR 2021)},
address = {Santiago de Chile (Virtual)},
url = {https://oceania.inria.cl/#aimocc},
editor = {Sanchez-Pi, Nayat and Mart\'{i}, Luis}
}
Online preprints
de Wolff, T., Carrillo, H., Martí, L., & Sanchez-Pi, N. (2021, May). Towards Optimally Weighted Physics-Informed Neural Networks in Ocean Modelling.
hal: hal-03260357pdfbibtex
@unpublished{dewolff:hal-03260357,
title = {Towards Optimally Weighted Physics-Informed Neural Networks in Ocean Modelling},
author = {de Wolff, Taco and Carrillo, Hugo and Mart\'{i}, Luis and Sanchez-Pi, Nayat},
year = {2021},
month = may,
url = {https://hal.inria.fr/hal-03260357},
hal_id = {hal-03260357},
hal_version = {v1},
pdf = {https://hal.inria.fr/hal-03260357/file/neurips_neural-based_solver.pdf},
eprint = {2106.08747},
archiveprefix = {arXiv},
primaryclass = {physics.ao-ph}
}
The Anthropocene has brought along a drastic impact on almost all life forms on the planet. Considering the importance and amount of water in this speck of dust in the middle of nowhere that we inhabit, we should have called it Planet Ocean. Oceans are not only important because of their volume but are also about the functions and contributions they provide to biodiversity, the human species included.
The goal of this workshop is to bring together researchers that are interested and/or applying AI and ML techniques to problems related to marine biology, modeling, and climate change mitigation. We also expect to attract natural science researchers interested in learning about and applying modern AI and ML methods. Consequently, the workshop will be a first stone on building a multi-disciplinary community behind this research topic, with collaborating researchers that share problems, insights, code, data, benchmarks, training pipelines, etc. Together, we aim to ultimately address an urgent matter regarding the future of humankind, nature, and our planet.
Workshop programme
The workshop will take place on Friday, 7 May 2021. A zoom link will be shared to allow the participation anyone insterested.
Papers and presentations will be made available ASAP. Please note that programme times are in the CLT/EST (UTC-5h)
09:00 - 09:05. Opening comments and welcome by the organizers.
09:05 - 09:45. Keynote presentation: Jacques Sainte-Marie, ANGE Team (Inria Paris and Sorbonne Université).
09:45 - 10:05. Investigating Ground-level Ozone Formation: A Case Study in Taiwan. Yu-Wen Chen (Academia Sinica), Sourav Medya (Northwestern University), and Yi-Chun Chen (Academia Sinica). abstractpaper (pdf)
Tropospheric ozone (O3) is a greenhouse gas which can absorb heat and make the weather even hotter during extreme heatwaves. Besides, it is an influential ground-level air pollutant which can severely damage the environment. Thus evaluating the importance of various factors related to the O3 formation process is essential. However, O3 simulated by the available climate models exhibits large variance in different places, indicating the insufficiency of models in explaining the O3 formation process correctly. In this paper, we aim to identify and understand the impact of various factors on O3 formation and predict the O3 concentrations under different pollution-reduced and climate change scenarios. We employ six supervised methods to estimate the observed O3 using fourteen meteorological and chemical variables. We find that the deep neural network (DNN) and long short-term memory (LSTM) based models can predict O3 concentrations accurately. We also demonstrate the importance of several variables in this prediction task. The results suggest that while Nitrogen Oxides negatively contributes to predicting O3, solar radiation makes a significantly positive contribution. Furthermore, we apply our two best models on O3 prediction under different global warming and pollution reduction scenarios to improve the policy-making decisions in the O3 reduction.
10:05 - 10:25. Model Discovery in the Sparse Sampling Regime. Gert-Jan Both, Georges Tod, and Remy Kusters (CRI). abstractpaper (pdf)
To improve the physical understanding and the predictions of complex dynamic systems, such as ocean dynamics and weather predictions, it is of paramount interest to identify interpretable models from coarsely and off-grid sampled observations. In this work we investigate how deep learning can improve model discovery of partial differential equations when the spacing between sensors is large and the samples are not placed on a grid. We show how leveraging physics informed neural network interpolation and automatic differentiation, allow to better fit the data and its spatiotemporal derivatives, compared to more classic spline interpolation and numerical differentiation techniques. As a result, deep learning based model discovery allows to recover the underlying equations, even when sensors are placed further apart than the data’s characteristic length scale and in the presence of high noise levels. We illustrate our claims on both synthetic and experimental data sets where combinations of physical processes such as (non)-linear advection, reaction and diffusion are correctly identified.
10:25 - 10:45. Physically-Consistent Generative Adversarial Networks for Coastal Flood Visualization. Björn Lütjens (MIT), Brandon Leshchinskiy (MIT), Christian Requena-Mesa (Computer Vision Group, Friedrich Schiller University Jena; DLR Institute of Data Science, Jena; Max Planck Institute for Biogeochemistry, Jena), Farrukh Chishtie (Spatial Informatics Group), Natalia Diaz Rodriguez (ENSTA Paris and INRIA Flowers), Oceane Boulais (NOAA), Aruna Sankaranarayanan (MIT), Aaron Piña (NASA Headquarters), Yarin Gal (University of Oxford), Chedy Raissi (INRIA), Alexander Lavin (Institute for Simulation Intelligence), and Dava Newman (MIT). abstractpaper (pdf)
As climate change increases the intensity of natural disasters, society needs better tools for adaptation. Floods, for example, are the most frequent natural disaster, and better tools for flood risk communication could increase the support for flood-resilient infrastructure development. Our work aims to enable more visual communication of large-scale climate impacts via visualizing the output of coastal flood models as satellite imagery. We propose the first deep learning pipeline to ensure physical-consistency in synthetic visual satellite imagery. We advanced a state-of-the-art GAN called pix2pixHD, such that it produces imagery that is physically-consistent with the output of an expert-validated storm surge model (NOAA SLOSH). By evaluating the imagery relative to physics-based flood maps, we find that our proposed framework outperforms baseline models in both physical-consistency and photorealism. We envision our work to be the first step towards a global visualization of how climate change shapes our landscape. Continuing on this path, we show that the proposed pipeline generalizes to visualize arctic sea ice melt. We also publish a dataset of over 25k labelled image-pairs to study image-to-image translation in Earth observation.
10:45 - 11:05. Coffee break and short paper discussions.
Short papers:
PCE-PINNs: Physics-Informed Neural Networks for Uncertainty Propagation in Ocean Modeling. Björn Lütjens (MIT), Mark Veillette (MIT Lincoln Laboratory), Dava Newman (MIT), and Cait Crawford (IBM). abstract
Climate models project an uncertainty range of possible warming scenarios from 1.5 to 5°C global temperature increase until 2100, according to the CMIP6 model ensemble. Climate risk management and infrastructure adaptation requires the accurate quantification of the uncertainties at the local level. Ensembles of high-resolution climate models could accurately quantify the uncertainties, but most physics-based climate models are computationally too expensive to run as ensemble. Recent works in physics-informed neural networks (PINNs) have combined deep learning and the physical sciences to learn up to 15k faster copies of climate submodels. However, the application of PINNs in climate modeling has so far been mostly limited to deterministic models. We reformulate a novel technique to combine polynomial chaos expansion (PCE), a classic technique for uncertainty propagation, with PINNs. The proposed PCE-PINNs learn a fast surrogate model that is demonstrated for uncertainty propagation of known parameter uncertainties. We showcase the effectiveness in ocean modeling by using the local advection-diffusion equation.
Generative modeling of spatio-temporal weather patterns with extreme event conditioning. Konstantin Klemmer (University of Warwick), Sudipan Saha (Technical University of Munich), Matthias Kahl (Technical University of Munich), Tianlin Xu (London School of Economics and Political Science), and Xiaoxiang Zhu (Technical University of Munich). abstractpaper (pdf)
Deep generative models are increasingly used to gain insights in the geospatial data domain, e.g., for climate data. However, most existing approaches work with temporal snapshots or assume 1D time-series; few are able to capture spatio-temporal processes simultaneously. Beyond this, Earth-systems data often exhibit highly irregular and complex patterns, for example caused by extreme weather events. Because of climate change, these phenomena are only increasing in frequency. Here, we proposed a novel GAN-based approach for generating spatio-temporal weather patterns conditioned on detected extreme events. Our approach augments GAN generator and discriminator with an encoded extreme weather event segmentation mask. These segmentation masks can be created from raw input using existing event detection frameworks. As such, our approach is highly modular and can be combined with custom GAN architectures. We highlight the applicability of our proposed approach in experiments with real-world surface radiation and zonal wind data.
CropGym: A reinforcement learning environment for crop management. Hiske Overweg, Herman Berghuijs, and Ioannis N. Athanasiadis (Wageningen University and Research). abstractpaper (pdf)
Nitrogen fertilizers have a detrimental effect on the environment, which can be reduced by optimizing fertilizer management strategies. We implement an OpenAI Gym environment where a reinforcement learning agent can learn fertilization management policies using process-based crop growth models and identify policies with reduced environmental impact. In our environment, an agent trained with the Proximal Policy Optimization algorithm is more successful at reducing environmental impacts than the other baseline agents we present.
Frost Forecasting Model using Graph Neural Networks with Spatio-Temporal Attention Hernán Lira, Luis Martí, and Nayat Sanchez-Pi (Inria Chile Research Center). abstractpaper (pdf)
Frost forecast is an important issue in climate research because of its economical impact in several industries. In this study a graph neural network (GNN) with spatio-temporal architecture is proposed to predict minimum temperatures in a experimental site. The model consider spatial and temporal relations and process multiple time series simultaneously. Performing predictions of 6, 12 and 24 hrs this model outperforms statistical and non-graph deep learning models.
11:45 - 12:05. Feature Importance in a Deep Learning Climate Emulator. Wei Xu (Brookhaven National Laboratory), Xihaier Luo (Brookhaven National Laboratory), Yihui (Ray) Ren (Brookhaven National Laboratory), Ji Hwan Park (Brookhaven National Laboratory), Shinjae Yoo (Brookhaven National Laboratory), and Balu Nadiga (Los Alamos National Lab). abstractpaper (pdf)
We present a study using a class of post-hoc local explanation methods i.e., feature importance methods for “understanding” a deep learning (DL) emulator of climate. Specifically, we consider a multiple-input-single-output emulator that uses a DenseNet encoder-decoder architecture and is trained to predict interannual variations of sea surface temperature (SST) at 1, 6, and 9 month lead times using the preceding 36 months of (appropriately filtered) SST data. First, feature importance methods are employed for individual predictions to spatio-temporally identify input features that are important for model prediction at chosen geographical regions and chosen prediction lead times. In a second step, we also examine the behavior of feature importance in a generalized sense by considering an aggregation of the importance heatmaps over training samples. We find that: 1) the climate emulator’s prediction at any geographical location depends dominantly on a small neighborhood around it; 2) the longer the prediction lead time, the further back the “importance” extends; and 3) to leading order, the temporal decay of“importance” is independent of geographical location. An ablation experiment is adopted to verify the findings. From the perspective of climate dynamics, these findings suggest a dominant role for local processes and a negligible role for re-mote teleconnections at the spatial and temporal scales we consider. From the perspective of network architecture, the spatio-temporal relations between the inputs and outputs we find suggest potential model refinements. We discuss further extensions of our methods, some of which we are considering in ongoing work.
12:05 - 12:25. Assessing Physics Informed Neural Networks in Ocean Modelling and Climate Change Applications. Taco de Wolff, Hugo Carrillo Lincopi, Luis Martí, and Nayat Sanchez-Pi (Inria Chile Research Center). abstractpaper (pdf)
The carbon pump of the world’s oceans plays a vital role in the biosphere and climate of the earth, urging improved understanding of the functions and influences of the oceans for climate change analyses. State-of-the-art techniques are required to develop models that can capture the complexity of ocean currents and temperature flows. We will explore the benefits of using physics informed neural networks (PINNs) for solving partial differential equations related to ocean modeling, such as the wave, shallow water, and advection-diffusion equations. PINNs account for adherence to physical laws in order to improve learning and generalization. However, in this work, we show that we observe worse training and generalizability results, contrary to recent publications.
12:25 - 12:45. Deep Embedded Clustering for BioAcoustic Clustering of Marine Mammal Vocalization. Ali Jahangirnezhad (University of Washington Bothell) and Afra Mashhadi (University of Washington). abstractpaper (pdf)
With the decrease of hardware costs, stationary hydrophones are increasingly deployed in the marine environment to record animal vocalizations amidst ocean noise over an extended period of time. Bioacoustic data collected in this way is an important and practical source to study vocally active marine species and can make an important contribution to ecosystem monitoring. However, a main challenge of this data is the lack of annotation which many supervised neural network models rely on to learn to distinguish between noise and marine animal vocalizations. In this paper, we posit an unsupervised deep embedded clustering based on LSTM autoencoders, that aims to learn the representation of the input audio by minimizing the reconstruction loss and to simultaneously minimize a clustering loss through Kullback–Leibler divergence.
12:45 - 13:20. Keynote presentation: Michèle Sebag, TAU Team (LISN, Inria, CNRS, and Univ. Paris Saclay).
We welcome submissions of long (8 pages) full papers and short (4 pages) summary papers. To prepare your submission, please use the ICLR 2021 LaTeX style files provided at: https://github.com/ICLR/Master-Template. Use the following link to submit your proposal(s): AIMOCC 2021 CMT submission site.
Important dates
[Updated] Submission deadline : April 12, 2021 (UTC-12).
Notification of acceptance: April 19, 2021.
[Updated] Reception of final version: May 5, 2021.
Topics
Topics of interest of this workshop can be grouped into two sets:
Addressing and advancing the state of the art in areas like AI, ML, mathematical modeling and simulation. Here the focus is set on:
improving neural network handling of graph-structured information,
improving the capacity of ML methods to learn in small data contexts,
understanding causal relations, interpretability and explainability in AI,
integrating model-driven and data-driven approaches, and
to develop, calibrate, and validate existing mechanistic models.
Focus on answering the questions from the application domain, where the main questions to be addressed are:
Which are the major patterns in plankton taxa and functional diversity?
Which are the major drivers of patterns and how do they interact?
How these patterns and drivers will likely change under climate change?
How will these changes affect the capacity of ocean ecosystems to sequester carbon from the atmosphere, that is the biological carbon pump?
What relations bind communities and local conditions?
What are the links between biodiversity functioning and structure?
How modern AI and computer vision can be applied as research and discovery support tools to understand planktonic communities?
How new biological knowledge can be derived from the application of anomaly detection, causal learning, and explainable AI.
Organizers
Nayat Sánchez-Pi and Luis Martí, Inria Chile.
Scientific committee
José Manuel Molina, Universidad Carlos III de Madrid,
Julien Salomon and Jacques Sainte-Marie, Inria Paris,
Olivier Bernard, Inria Sophia-Antipolis,
Michèle Sebag and Marc Schoenauer, Inria Saclay,
Alejandro Maass, Center of Mathematical Modeling (CMM), Universidad de Chile,
Pablo Marquet, Pontificia Universidad Católica de Chile (PUC),
André Abreu, Fondation TARA Océan,
Ana Cristina Garcia, Unirio - Federal University of Rio de Janeiro State,
Hernán Lira, Inria Chile Research Center,
Hugo Carrillo Lincopi, Inria Chile Research Center,
Leandro Fernandes, Universidade Federal Fluminense,
Roberto Santana, University of the Basque Country (UPV/EHU),
Colomban De Vargas, GO-SEE CNRS Federation, and
Damien Eveillard, ComBi, Université de Nantes.
Diversity commitment
We will seek diversity in all aspects, both in school of thought, nationalities, stages in the academic career, etc.
Access
We will publish the accepted papers and talk abstracts (before the event) and the slides of the speakers (after the event) on the workshop website. We will include a bibliography of most relevant research papers to facilitate cross pollination of ideas between these fields. Similarly, we will record the workshop and publish it online.
The Anthropocene has brought along a drastic impact on almost all life forms on the planet. Considering the importance and amount of water in this speck of dust in the middle of nowhere that we inhabit, we should have called it Planet Ocean. Oceans are not only important because of their volume but are also about the functions and contributions they provide to biodiversity, the human species included.
The goal of this workshop is to bring together researchers that are interested and/or applying AI and ML techniques to problems related to marine biology, modeling, and climate change mitigation. We also expect to attract natural science researchers interested in learning about and applying modern AI and ML methods. Consequently, the workshop will be a first stone on building a multi-disciplinary community behind this research topic, with collaborating researchers that share problems, insights, code, data, benchmarks, training pipelines, etc. Together, we aim to ultimately address an urgent matter regarding the future of humankind, nature, and our planet.
14:00 - 14:15. Opening comments and welcome by the organizers.
14:15 - 14:40. A Physics-Informed Neural Network to Model Port Channels. Marlon S. Mathias1, Caio Fabricio Deberaldini Netto1, Marcel M.B. Barros1, Jefferson F. Coelho2, Lucas P. de Freitas1, Felipe M. Moreno1, Fabio Cozman1, Anna Helena Reali Costa 1, Eduardo Aoun Tannuri1, Edson S. Gomi 1, and Marcelo Dottori3. (1) University of São Paulo, (2) São Paulo University (POLI-USP), (3) Oceanographic Institute, University of São Paulo. abstractpaper (pdf)online presentation
We describe a Physics-Informed Neural Network (PINN) that simulates the flow induced by the astronomical tide in a synthetic port channel, with dimensions based on the Santos-São Vicente-Bertioga Estuarine System. PINN models aim to combine the knowledge of physical systems and data-driven machine learning models. This is done by training a neural network to minimize the residuals of the governing equations in sample points. In this work, our flow is governed by the Navier-Stokes equations with some approximations. There are two main novelties in this paper. First, we design our model to assume that the flow is periodic in time, which is not feasible in conventional simulation methods. Second, we evaluate the benefit of resampling the function evaluation points during training, which has a near zero computational cost and has been verified to improve the final model, especially for small batch sizes. Finally, we discuss some limitations of the approximations used in the Navier-Stokes equations regarding the modeling of turbulence and how it interacts with PINNs.
14:40 - 15:05. Towards Optimally Weighted Physics-Informed Neural Networks in Ocean Modelling. Hugo Carrillo Lincopi, Taco de Wolff, Luis Martí, and Nayat Sánchez Pi. Inria Chile Research Center. abstractpaper (pdf)online presentation
Understanding the ocean has particular relevance with the emergence of the climate change phenomenon. Nowadays, this is an essential task, but also very expensive in the computational sense. This work explores the benefits of using physics-informed neural networks (PINNs) for solving partial differential equations (PDEs) related to ocean modeling; such as the Burgers, wave, and advection-diffusion equations. We explore the trade-offs of using data vs. physical models in PINNs for solving partial differential equations. PINNs account for the deviation from physical laws in order to improve learning and generalization. We observed how the relative weight between the data and physical model in the loss function influences training results. Additionally, we compare the variance of our results and analyze the implications of activation functions for training neural network derivatives.
15:05 - 15:30. Modeling Oceanic Variables with Dynamic Graph Neural Networks. Caio Fabricio Deberaldini Netto1, Marcel M.B. Barros1, Jefferson F. Coelho2, Felipe M. Moreno1, Marlon S. Mathias1, Lucas P. de Freitas1, Fabio Cozman1, Marcelo Dottori3, Eduardo Aoun Tannuri1, Edson S. Gomi1, and Anna Helena Reali Costa1. (1) University of São Paulo, (2) São Paulo University (POLI-USP), (3) Oceanographic Institute, University of São Paulo. abstractpaper (pdf)online presentation
Researchers typically resort to numerical methods to understand and predict ocean dynamics, a key task in mastering environmental phenomena. Such methods may not be suitable in scenarios where the topographic map is complex, knowledge about the underlying processes is incomplete, or the application is time critical. On the other hand, if ocean dynamics are observed, they can be exploited by recent machine learning methods. In this paper we describe a data-driven method to predict environmental variables such as current velocity and sea surface height in the region of Santos-Sao Vicente-Bertioga Estuarine System in the southeastern coast of Brazil. Our model exploits both temporal and spatial inductive biases by joining state-of-the-art sequence models (LSTM and Transformers) and relational models (Graph Neural Networks) in an end-to-end framework that learns both the temporal features and the spatial relationship shared among observation sites. We compare our results with the Santos Operational Forecasting System (SOFS). Experiments show that better results are attained by our model, while maintaining flexibility and little domain knowledge dependency.
15:30 - 16:00. Coffee break (we stay in the online call and chat).
16:00 - 16:25. Enhancing Oceanic Variables Forecast in the Santos Channel by Estimating Model Error with Random Forests. Felipe M. Moreno1, Caio Fabricio Deberaldini Netto1, Marcel M.B. Barros1, Jefferson F. Coelho2, Lucas P de Freitas1, Marlon S. Mathias1, Luiz Schiaveto Neto3, Marcelo Dottori4, Fabio Cozman1, Anna Helena Reali Costa1, Edson S. Gomi1, and Eduardo Aoun Tannuri 1. (1) University of São Paulo, (2) São Paulo University (POLI-USP), (3) Escola Politécnica – University of Sao Paulo, (4) Oceanographic Institute, University of São Paulo. abstractpaper (pdf)online presentation
In this work we improve forecasting of Sea Surface Height (SSH) and current velocity (speed and direction) in oceanic scenarios. We do so by resorting to Random Forests so as to predict the error of a numerical forecasting system developed for the Santos Channel in Brazil. We have used the Santos Operational Forecasting System (SOFS) and data collected in situ between the years of 2019 and 2021. In previous studies we have applied similar methods for current velocity in the channel entrance, in this work we expand the application to improve the SHH forecast and include four other stations in the channel. We have obtained an average reduction of 11.9% in forecasting Root-Mean Square Error (RMSE) and 38.7% in bias with our approach. We also obtained an increase of Agreement (IOA) in 10 of the 14 combinations of forecasted variables and stations.
16:25 - 16:50. The BLue Amazon Brain (BLAB): A Modular Architecture of Services about the Brazilian Maritime Territory. Paulo Pirozelli1, Ais B.R. Castro1, Ana Luiza C. de Oliveira1, André Seidel1, Flávio N. Cação1, Igor C. Silveira1, João G M Campos1, Laura C. Motheo1, Leticia F. Figueiredo1, Lucas F.A.O. Pellicer1, Marcelo A. José1, Marcos M. José1, Pedro de M. Ligabue1, Ricardo S. Grava1, Rodrigo M. Tavares1, Vinícius B. Matos1, Yan V. Sym1, Anna Helena Reali Costa1, Anarosa Alves Franco Brandão2, Denis D. Maua1 Fabio Cozman1, Sarajane M. Peres1. (1) University of São Paulo, (2) Escola Politécnica – University of Sao Paulo. abstractpaper (pdf)online presentation
We describe the first steps in the development of an artificial agent focused on the Brazilian maritime territory, a large region within the South Atlantic also known as the Blue Amazon. The “BLue Amazon Brain” (BLAB) integrates a number of services aimed at disseminating information about this region and its importance, functioning as a tool for environmental awareness. The main service provided by BLAB is a conversational facility that deals with complex questions about the Blue Amazon, called BLAB-Chat; its central component is a controller that manages several task-oriented natural language processing modules (e.g., question answering and summarizer systems). These modules have access to an internal data lake as well as to third-party databases. A news reporter (BLAB-Reporter) and a purposely-developed wiki (BLAB-Wiki) are also part of the BLAB service arquitecture. In this paper, we describe our current version of BLAB’s architecture (interface, backend, web services, NLP modules, and resources) and comment on the challenges we have faced so far, such as the lack of training data and the scattered state of domain information. Solving these issues presents a considerable challenge in the development of artificial intelligence for technical domains.
16:50 - 17:00. Final remarks.
17:00 - until available. Open topic conversations.
Submissions
We welcome submissions of full papers (8 pages, not counting references) and short summary papers (4 pages, not counting references). Papers must be written in English and in PDF format according to the IJCAI-ECAI’22 style. All submitted papers will be under a single-blinded peer review for their novelty, technical quality and impact. The submissions can contain author details.
We will seek to publish selected, revised, extended papers later in a planned post-proceedings volume, to be published in the Lecture Notes in Artificial Intelligence (LNAI) series. The selection of papers will be managed by a subset of the workshop organizing committee.
José Manuel Molina, Universidad Carlos III de Madrid,
Julien Salomon, ANGE, Inria Paris,
Jacques Sainte-Marie, ANGE, Inria Paris,
Olivier Bernard, BIOCORE, Inria Sophia-Antipolis,
Michèle Sebag, TAU, Inria Saclay,
Marc Schoenauer, TAU, Inria Saclay,
Pablo Marquet, Pontificia Universidad Católica de Chile (PUC),
André Abreu, Fondation TARA Océan,
Ana Cristina Garcia Bicharra, Unirio - Federal University of Rio de Janeiro State,
Hernán Lira, Inria Chile Research Center,
Hugo Carrillo Lincopi, Inria Chile Research Center,
Andrew Berry, Inria Chile Research Center,
Luis Valenzuela, Inria Chile Research Center,
Leandro Fernandes, Universidade Federal Fluminense,
Roberto Santana, University of the Basque Country (UPV/EHU),
Colomban De Vargas, GO-SEE CNRS Federation, and
Damien Eveillard, ComBi, Université de Nantes.
Diversity commitment
We will seek diversity in all aspects, both in school of thought, nationalities, stages in the academic career, etc.
Access
We will publish the accepted papers and talk abstracts (before the event) and the slides of the speakers (after the event) on the workshop website. We will include a bibliography of most relevant research papers to facilitate cross-pollination of ideas between these fields. Similarly, we will record the workshop and publish it online.
OcéanIA IJCAI 2022 Challenge
AI methods for determining ocean ecosystems from space: Combining genomic information, microscopic and satellite imagery
The ocean is the Earth’s principal climate regulator and the main responsible for sequestering carbon dioxide (CO2). This makes it our main defense against climate change, but climate change itself is destroying the healing capacity of the ocean. Algae and, in particular, plankton, play a fundamental role in this, as they are able to remove CO2. Therefore, the mitigating capacity of an ecosystem can be established based on the presence of particular types of plankton. However, to health of the larger areas of the ocean can only be determined through large-scale measurements such as satellite imagery.
The challenge focuses on the remote identification via satellite imagery of high-potential ecosystems. This would allow large tracts of the ocean to be analyzed in a way that allows scientists and decision makers to understand how the ocean evolves over time and could be used to create policies for protecting high-value parts of the ocean. Alternatively, we propose to study the use of marker species, such as whales, which can be identified and their presence implies the existence of others.
This is an opportunity to attract the AI/ML community to this type of scientifically challenging and high-impact problem. For this we will make available to participants curated georeferenced datasets of plankton images, genomic data and satellite images and provide mentorship during the period of the challenge. It falls under the activities of Inria Project OcéanIA.
Goals
We propose to determine the variation of plankton species —i.e. ecosystems— inhabiting a given area of the ocean by cross-referencing genomic data, plankton microscope imaging and satellite images. This calls for the combined application of methods like:
causal inference,
explainable AI,
computer vision neural networks: representation learning, self-supervision, out of distribution detection,
ML methods for “small data contexts” like zero-shot/few-shot learning, and active learning, among others,
associative rule learning, and
domain adaptation and transfer learning, to mention a few.
Participation guide
The challenge will take place from 20 April 20 2022 to 29 July 2022. Teams can join the challenge at any time, but we suggest you that you do it as early as possible.
The challenge is organized in two phases:
Phase I: where participants work on a solution proposal and plan.
At the end of this phase, participants must submit a short paper (max. 2 pages excl. references) and (optionally) supplementary code.
Submitted proposals are evaluated, and selected ones are invited to take part of the phase II.
Phase II: where participants work on their challenge solutions.
At the end of this phase, participants should submit a full paper (max. 6 pages excl. references) and make available the supplementary code under an OSI approved license (i.e. MIT, Apache, etc.).
Bring your own data: we encourage participants to make available additional datasets. Share them under an open license and follow the contribution instructions above to add them to the guide.
Getting involved
Join the mailing list: If you are interested to take part of the challenge, please let us know by filling up this form.
Join out discord server to get support, collaborate and exchange with other participants.
Papers must be written in English and in PDF format according to the IJCAI-ECAI’22 style. All submitted papers will be under a single-blinded peer review for their novelty, technical quality and impact. The submissions can contain author details. See below for submission link.
Source code instructions
The challenge will help bring recent state-of-the-art AI/ML methods to tackle complex and high-impact problems that have a potential for global impact. Experts on this field have limited access and operational knowledge on how to use these advanced methods. Consequently, we will pay extra attention and involve participants in order to make their code contributions available in a form as usable as possible by non-AI/ML experts.
During the unfolding of the challenge source code availability (open source or private) will be left to the decision of the participants.
Derived and/or intermediate datasets that we consider of value will also be made freely available.
Upon acceptance, participants code should be made available online under an open-source friendly license, in particular it should be an OSI approved license.
We encourage participants to make their source code as easy to use as possible by providing installation scripts, instructions, etc.
Phase I. Solution proposal preparation (20 April – 7 June 2022).
During this phase participants work on the conception of their solutions.
Participants are encouraged to interact via email or discord with organizers and other participants.
Submission of proposals (7 June 2022, UTC-12). Proposal submissions must include:
Short paper (max 2 pages excluding references) with the proposed solution, potential impact, planning, etc., and
Code repo (optional) link to code repository (i.e. GitHub, GitLab, etc.) with supplementary code. This code does not need to be public, but in that case organizers should have access granted to it.
Challenge session and awards at IJCAI-ECAI’22 (July 22-29, 2022): Participants will present their solutions in an in-person session in the conference. Only participants registered at ICJAI-ECAI’22 will be able to take part of the session.
Awards
We will provide small ocean-related gifts and cloud compute to the best contributions. Stay tuned for more details.
Publications and Post-proceedings
Dissemination is very important for the goals of the challenge. We will publish a non-archival proceedings booklet with the contributions and the main experiences gained during the challenge. Therefore, both the peer-review post volume and the challenge paper describing the results, experiences and lessons learned are interesting for us.
José Manuel Molina, Universidad Carlos III de Madrid,
Pablo Marquet, Pontificia Universidad Católica de Chile.
Julien Salomon, ANGE, Inria Paris,
Jacques Sainte-Marie, ANGE, Inria Paris,
Olivier Bernard, BIOCORE, Inria Sophia-Antipolis,
Michèle Sebag, TAU, Inria Saclay,
Marc Schoenauer, TAU, Inria Saclay,
Alejandro Maass, Center of Mathematical Modeling (CMM), Universidad de Chile,
Pablo Marquet, Pontificia Universidad Católica de Chile (PUC),
André Abreu, Fondation TARA Océan,
Ana Cristina Garcia Bicharra, Unirio - Federal University of Rio de Janeiro State,
Hernán Lira, Inria Chile Research Center,
Hugo Carrillo Lincopi, Inria Chile Research Center,
Leandro Fernandes, Universidade Federal Fluminense,
Roberto Santana, University of the Basque Country (UPV/EHU),
Colomban De Vargas, GO-SEE CNRS Federation, and
Damien Eveillard, ComBi, Université de Nantes.
Sponsorship
We are actively seeking support for different organizations. If you are interested to sponsor this challenge do not hesitate to contact us.
Diversity commitment
We will actively seek diversity in all aspects: schools of thought, theoretical backgrounds, nationalities, stages in the academic career, gender, etc. We will take an affirmative action to ensure that by disseminating the call for papers in diverse communities and offer a mentorship and assistance to help underrepresented and cross-disciplinary participants.
The Anthropocene has brought along a drastic impact on almost all life forms on the planet. Considering the importance and amount of water in this speck of dust in the middle of nowhere that we inhabit, we should have called it Planet Ocean. Oceans are not only important because of their volume but are also about the functions and contributions they provide to biodiversity, the human species included.
The goal of this workshop is to bring together researchers that are interested and/or applying AI and ML techniques to problems related to marine biology, modeling, and climate change mitigation. We also expect to attract natural science researchers interested in learning about and applying modern AI and ML methods. Consequently, the workshop will be a first stone on building a multi-disciplinary community behind this research topic, with collaborating researchers that share problems, insights, code, data, benchmarks, training pipelines, etc. Together, we aim to ultimately address an urgent matter regarding the future of humankind, nature, and our planet.
Topics
Topics of interest of this workshop can be grouped into two sets:
Addressing and advancing the state of the art in areas like AI, ML, mathematical modeling and simulation. Here the focus is set on:
improving neural network handling of graph-structured information,
improving the capacity of ML methods to learn in small data contexts,
understanding causal relations, interpretability and explainability in AI,
integrating model-driven and data-driven approaches, and
to develop, calibrate, and validate existing mechanistic models.
Focus on answering the questions from the application domain, where the main questions to be addressed are:
Which are the major patterns in plankton taxa and functional diversity?
Which are the major drivers of patterns, and how do they interact?
How these patterns and drivers will likely change under climate change?
How will these changes affect the capacity of ocean ecosystems to sequester carbon from the atmosphere, that is the biological carbon pump?
What relations bind communities and local conditions?
What are the links between biodiversity functioning and structure?
How modern AI and computer vision can be applied as research and discovery support tools to understand planktonic communities?
How new biological knowledge can be derived from the application of anomaly detection, causal learning, and explainable AI.
Submissions
We welcome submissions of full papers (8 pages, not counting references) and short summary papers (4 pages, not counting references). Papers must be written in English and in PDF format according to the ECAI’24 LaTeX template. All submitted papers will be under a single-blinded peer review for their novelty, technical quality and impact. The submissions can contain author details.
We will seek to publish selected, revised, extended papers later in a planned post-proceedings volume, to be published in the Lecture Notes in Artificial Intelligence (LNAI) series. The selection of papers will be managed by a subset of the workshop organizing committee.
José Manuel Molina, Universidad Carlos III de Madrid,
Julien Salomon, ANGE, Inria Paris,
Jacques Sainte-Marie, ANGE, Inria Paris,
Olivier Bernard, BIOCORE, Inria Sophia-Antipolis,
Michèle Sebag, TAU, Inria Saclay,
Marc Schoenauer, TAU, Inria Saclay,
Pablo Marquet, Pontificia Universidad Católica de Chile (PUC),
André Abreu, Fondation TARA Océan,
Ana Cristina Garcia Bicharra, Unirio - Federal University of Rio de Janeiro State,
Hernán Lira, Inria Chile Research Center,
Luis Valenzuela, Inria Chile Research Center,
Leandro Fernandes, Universidade Federal Fluminense,
Roberto Santana, University of the Basque Country (UPV/EHU),
Colomban De Vargas, GO-SEE CNRS Federation, and
Damien Eveillard, ComBi, Université de Nantes.
Diversity commitment
We will seek diversity in all aspects, both in school of thought, nationalities, stages in the academic career, etc.
Access
We will publish the accepted papers and talk abstracts (before the event) and the slides of the speakers (after the event) on the workshop website. We will include a bibliography of most relevant research papers to facilitate cross-pollination of ideas between these fields. Similarly, we will record the workshop and publish it online.
Software
Extract biologic subsequences of interest from large FASTA files
Serverless cloud service. Focus on your query, not on managing storage or compute infraestructure.
Preloaded data catalog. No need to move large files around.
Access the service right from your Python code. Get query results as a Pandas DataFrame.
Run a sample query on a Jupyter notebook now!
How does it work?
OcéanIA Platform lets you query large FASTA files in our supported data catalogs for extracting parts of biologic sequences. Just import our Python library in your code and access the query service.
Run your queries locally on your workstation either through a Python script or a Jupyter notebook. Install our Python package, then import our Python module in your code, and you are ready to go.
QUERY FASTA FILES
Extract multiple specific gene sequences from a file in a single query.
How to use
Run pip install oceania-query-fasta to install the Python package that enables access to query service.
From your Python code (either a script or a notebook) import the oceania module.
Select a file from the catalog, define your query, apply the query to the file. The result can be either retrieved as a Pandas DataFrame or saved as a CSV file.
A Jupyter notebook with a minimal example of a query that extracts a set of sequences from a given file in the catalog.
A Jupyter notebook with an example that extracts large intergenic regions (IGRs) from one surface sample.
#OcéanIA🌊aporta a la comprensión del océano, desarrollando herramientas de la #IA que contribuyan a la comprensión de la estructura, funcionamiento, mecanismos subyacentes y dinámicas del océano y su papel en la regulación de la biósfera y en la lucha contra el #cambioclimático.
Hoy en el Centro @inria_saclay los equipos que colaboran en el #InriaChallenge#OcéanIA🌊 realizaron la reunión anual en la que pusieron en común los avances del proyecto y conversaron acerca de sus resultados y próximos desafíos.
#OcéanIA🌊 es uno de los proyectos insignia de @Inria_Chile en este ámbito, que busca comprender el océano en sí mismo y su biodiversidad ya que es fundamental para la regulación del clima a escala global; y en el que se trabaja en colaboración con @CeodosChile y @TaraOcean_⛵.
Hacer ciencia en beneficio de la sociedad y el medio ambiente es nuestro compromiso, que se mantiene con #GreenAI y otros proyectos, como #OcéanIA🌊 , #Emistral⛵ , #FairTrees🌲 y #FrostForecast❄️, que abocan los esfuerzos de Inria Chile en la lucha contra el #cambioclimático.
Los datos colectados serán procesados en el contexto del proyecto científico-tecnológico de Inria Chile #OcéanIA que estudia el impacto del #cambioclimático y la relación entre el funcionamiento del ecosistema marino y la biodiversidad, usando diversas herramientas digitales.
@Inria_Chile estuvo presente hoy con el taller "#OceanIA: la #inteligenciaartificial para comprender los océanos y el cambio climático" a cargo de Luis Martí, director científico de @Inria_Chile y dirigida a una clase de 1º. del Liceo Jean Mermoz de Curicó.
De esto se trata nuestro Workshop "AI: Modeling Oceans and Climate Change (AIMOCC 2022)" que, en el marco de nuestro proyecto #OcéanIA, tendrá lugar hoy en @IJCAIconf#IJCAI2022#IA.
En el Whitebook "OcéanIA: IA, datos y modelos para comprender el océano y el cambio climático", podrás conocer en detalle este proyecto. Revísalo aquí 👉 https://t.co/Bxj7VJ1Otx
En esta ocasión, estuvimos presentes con #oceanIA dado su impacto en la determinación de la variación de las especies de plancton que habitan un área determinada del océano mediante el cruce de datos genómicos, imágenes de microscopio e imágenes de satélite.
You have until June 7, to apply for the challenge "AI METHODS FOR DETERMINING OCEAN ECOSYSTEMS FROM SPACE: COMBINING GENOMIC INFORMATION, MICROSCOPIC AND SATELLITE IMAGERY", which will take place in Vienna next July, within the activities of our #OceanIA 🌊 project in #IJCAI2022https://t.co/qSdpIm8mkR
He is researcher in the Inria Challenge #OcéanIA, our iconic project that seeks to develop new #AI and #MathematicalModeling tools to understand oceans' structures, functioning, underlying mechanisms and dynamics and its role in the regulation and preservation of the biosphere.
Es investigador del Inria Challenge #OcéanIA, nuestro proyecto ícono que busca desarrollar herramientas de #IA y #modeladomatemático que ayuden a entender la estructura, funcionamiento, mecanismos subyacentes y dinámica del océano y su rol en regular y mantener la biósfera.
We invite to all the AI/ML community to participate in the workshop "AI: Modeling Oceans and Climate Change" (AIMOCC 2022), organized by @inria_chile , that will take place within the IJCAI-ECAI 2022 in July in Vienna, Austria. pic.twitter.com/YUzh34E1GF
📢 ‼️ #IJCAI2022#OceanIA Participa del workshop "AI: Modeling Oceans and Climate Change" (AIMOCC 2022) organizado por @Inria_Chile y que se da en el marco del IJCAI-ECAI 2022, que se realizará entre el 23 y 29 de julio en Viena, Austria. pic.twitter.com/SHAsu0ZlY7
El desafío se da en el marco de en el marco de las actividades de nuestro proyecto #OcéanIA 🌊. ¿Te interesa? Apuntate al desafío aquí https://t.co/jZ1ipZr6oF
El desafío se da en el marco de en el marco de las actividades de nuestro proyecto #OcéanIA 🌊. ¿Te interesa? Apuntate al desafío aquí https://t.co/jZ1ipZr6oF
En Inria Chile estamos trabajando en esta materia (y seguiremos haciéndolo) con proyectos I+D, entre los que destacan: #OceanIA, para la comprensión de los océanos y cambio climático. #GreenAI, para reducir el impacto ecológico de la inteligencia artificial y el machine learning.
Continuamos en @FeteScience y hoy nuestro equipo realizó un taller para el Liceo Jean Mermoz de Curicó y el Liceo Jean d’Alembert de Viña del Mar, de la red AEFE Chile, sobre #OcéanIA y los efectos adversos del cambio climático en el océano global. pic.twitter.com/j73XNsgnZH
Ayer☝️nuestra directora @nayatsanchezpi presentó #OcéanIA en el evento "Cooperación Chile-Francia en Océanos, biodiversidad y cambio climático: Nuevas perspectivas científicas para explorar el océano" Mira su presentación aquí 👇🏻 y abajo extractos de ella https://t.co/SUJqhGdLPX
Te presentamos a @nayatsanchezpi, directora del Centro de Investigación de Inria en Chile. Ella compartirá con nosotros sobre "Los grandes desafíos del machine learning y combatir el cambio climático: una carretera de ida y vuelta".
Many thanks to @nayatsanchezpi (Director of @inria_chile) for taking the time to tell us about the OcéanIA project and this exciting and important research into oceans and climate. https://t.co/GoysKViprS
If you are attending #ICLR2021 come and join us in the workshop we organize #AIMOCC AI: MODELING OCEANS AND CLIMATE CHANGE 👉🏻https://t.co/8ZVqSVIq1F We invite you to follow the 3 excellent keynotes and 10 accepted papers.
Hoy, llegó a Valparaíso el velero de la Misión @TaraOcean_ , que estudia los microorganismos oceánicos. Somos parte de esta misión y lideramos #OcéanIA, proyecto que usará #IA para estudiar estos y otros datos y su impacto en el cambio climático. https://t.co/QcJy5D4XsNpic.twitter.com/5SkoHzxS7U
Alejandro Maass, director @CMMUChile, estuvo en el programa "Congreso Futuro" de @Cooperativa hablando del importante trabajo colaborativo hecho durante la última travesía del velero científico Tara por nuestro país. Podcast (desde el min 34:00) en: https://t.co/gLv2au4ld3pic.twitter.com/1hXQ3uWhNg
No sólo hay "Perseverancia" en la exploración de Marte. Los invito a seguir la expedición @TaraOcean_ ⛵️ y el laboratorio a bordo de CEODOS, un consorcio de 9 centros de excelencia nacionales para estudiar el impacto del cambio climático en nuestro océanohttps://t.co/JOW5rS9VeNpic.twitter.com/R7dl8f36nx
Nuestra extensa costa hace de Chile un laboratorio natural para el estudio de los océanos. Hoy llegó a nuestro país el velero científico Tara desde 🇫🇷 para analizar los efectos del cambio climático en el microbioma marino🌊. Respuestas desde🇨🇱 para el🌏. https://t.co/tfoFPFf9tx
¡Mañana! No te pierdas la transmisión especial de la llegada del velero TARA a Punta Arenas, marcando la primera parte de esta expedición internacional que recorrerá las costas de Chile, estudiando los efectos del cambio climático ¡Agéndalo!
— Departamento de Oceanografía UdeC (@DOCEUdeC) January 30, 2021
Los invitamos hoy a las 20:00 horas a conocer más sobre #OcéanIA en una entretenida conversación de @nayatsanchezpi con @humbertosichel en el programa 2040 de @RadioSonar. ¡Escúchala en la 105.3 FM!
En #Sonar2040@humbertosichel recibe a Nayat Sánchez-Pi, directora Ejecutiva de @inria_chile e investigadora líder de OceanIA, para conversar sobre los desafíos futuros en el área científica y de transferencia tecnológica.
¡Este proyecto nos tiene muy contentos! 🌊💻Se trata de #OcéanIA, un #InriaChallenge cuyo objetivo es comprender el cambio climático a través del estudio del océano usando inteligencia artificial.
“Oceans today we can say are the last unknown, and understanding the role of oceans in climate change is not only important but also a challenge for modern AI and applied ML." - @nayatsanchezpihttps://t.co/aLw5ifvRZW
Our last speaker will be Nayat Sanchez-Pi (@nayatsanchezpi) who will join us from Chile, where she directs the Inria Chile Research Center (@inria_chile). She will present her talk "OcéanIA: AI, Oceans and Climate Change".
The 2024 OcéanIA annual meeting took place on February 23, 2024, in the offices of Inria Paris.
The primary outcomes of the project so far can be summarized as:
Development of state-of-the-art computer vision techniques for explainable plankton identification, paving the way for a more cost-effective approach to describing oceanic ecosystems.
Identification of patterns linking genomic characteristics of ecosystems to their functions, enabling the recognition of valuable ecosystems on a global scale.
Advancements in modeling, physics-informed machine learning, and the coupling of complex systems models. These progressions aim to empower us to simulate climate change scenarios, predicting their impact on, and potential displacement of, plankton ecosystems.
The project’s diverse methodologies and interdisciplinary approaches have not only yielded significant findings but, perhaps more crucially, have laid the groundwork for an ambitious next phase.
Program
09:00 - 09:30 Welcome coffee.
09:30 - 09:40 Intro and words by Jean-Frédéric Gerbeau.
09:40 - 10:00 Nayat Sanchez and Luis Martí (Inria Chile): Overall status of OcéanIA, organization of the day and presentations. slides
10:00 - 10:30 Olivier Bernard and Francesca Casagli (BIOCORE): Effect of temperature and light on phytoplankton growth. slides
10:30 - 11:00 Romain Ranini (BIOCORE): Spatio-temporal high-resolution models of particulate organic matter abundance in the ocean. slides
11:30 - 12:00 M. Sebag, M. Schoenauer, and L. Martí (TAU + Inria Chile): Understanding ecosystems via plankton images. slides
12:00 - 11:30 Pierre Peterlongo (DYLISS): Invited presentation about Inria Challenge OmicsFinder.
12:30 - 13:00 Louis Thiry (ANGE): Simulation of several large-scale ocean models. slides
13:00 - 15:00 Lunch at Le Repaire (100 Av. Daumesnil 75012)
15:00 - 15:30 Luis Valenzuela (Inria Chile): Modeling global plankton communities via multiomics and ML approaches. slides
15:30 - 16:00 Pablo Marquet (PUC): A general theory for temperature dependence in biology. slides
16:00 - 16:30 Fabien Lombard and Jean-Olivier Irisson (Laboratoire d’Océanographie de Villefranche): Invited presentation about the Ecotaxa project.
16:30 - 16:45 Pause coffee.
16:45 - 17:10 Luis Martí (Inria Chile): Progress in Physics-informed ML and its applications in oceanic processes. slides
17:10 - 17:30 Speed presentations by Inria Chile interns of 2024.
17:30 - 18:00 Discussion of next steps and conclusions.
Project status notes
The project goal can be summarized as the pursuit of the answers to three increasingly complex interdependent questions: (i) how the Ocean mitigates climate change, (ii) how climate change will impact that capacity, and (iii) what can be done to protect the Ocean.
The overall organization of work in the project can be framed in a map/reduce schema. That is that in the first phase of the project (map) activities were focused on the specific research questions while in the second (reduce) phase would then focus on the convergence of the different results as results that can be shared and actionable.
Most of the activities reported in here fall under the map phase.
Understanding ecosystems via plankton images
We have made the first study on the application of state-of-the-art computer vision models for plankton identification. For this we have used massive raw, unprocessed large plankton datasets. We have also applied novel hierarchical classification, out-of-distribution detection methods, diversity-preserving loss functions, multi-modal/multi-task learning, strategies for handling class imbalance, and explainable AI methods to render the results more valuable.
We have also experimented with data augmentation and contrastive learning approaches to address the problem of identifying outliers (unknown species) using generalized contrastive losses and introducing fake outliers extracted from known images, or creating them as chimeras or applying generative AI methods.
Characterization of the Ocean carbon pump
The Ocean functions as a reservoir for CO2 through the biological carbon pump. This process is fueled by the metabolic diversity in plankton communities, which is encoded in their genetic material.
Such diversity is a cornerstone in fueling global biogeochemical cycles, underpinning food webs, and aiding in climate regulation.
Two main efforts have been made around this: (i) identification of relations between (meta) omic features with ecosystem functions and (ii) global scale spatio-temporal characterization of the carbon pump using the organic matter abundance.
In the first effort we focused on deciphering global patterns within the matrices derived from marine microbial datasets. To achieve this, we utilized unsupervised machine learning methods, such as uniform manifold approximation and projection (UMAP) and autoencoder neural networks.
Notably, samples from the mesopelagic zone clustered distinctly apart from other samples, and similarly, polar samples were clearly segregated from non-polar ones. We estimated Shapley additive explanations (SHAP) to understand clusters (polar vs. non-polar, mesopelagic vs. surface samples) to understand the factors driving them. We also applied novel principles like (i) topological data analysis, which assesses the topological features of data across various scales, (ii) symbolic regression, capable of identifying concise interpretable human-readable expressions directly from data, (iii) large language models for direct genomic tagging, and (iv) optimal transport to project metagenomic/metatranscriptomic characteristics across different depths.
We also focused on characterizing the different oceanic layers: surface, Deep Chlorophyll Maximum (DCM), and mesopelagic. For this purpose, we trained classification models and utilized SHAP (SHapley Additive exPlanations) values to dissect the variations in the metagenomic, metatranscriptomic, and taxonomic compositions across these layers. Our findings revealed that the primary distinctions between the top two layers (surface and DCM) are linked to aspects of carbon and amino acid metabolism, pathways of carbon fixation and, antimicrobial processes. In contrast, when analyzing the deeper mesopelagic layer, these processes appeared to be significantly different, being noteworthy processes involved in protein folding adaptations to high pressure and low temperature.
Symbolic regression, a cutting-edge machine learning technique that excels in identifying concise, interpretable human-readable expressions directly from data. This method was applied to explore the intricate relationships between environmental and biological variables. Specifically, we trained symbolic regression models to predict the environmental parameters, using features derived from metagenomic, metatranscriptomic, and taxonomic datasets as predictors. The results of this analysis were particularly striking for certain environmental variables. Our models achieved the highest R-squared (R2) coefficients in predicting oxygen levels and temperature. This was followed by accurate predictions for salinity, latitude, water density, and concentrations of silicate, nitrate, and phosphate. These findings underscore the potential of symbolic regression as a powerful tool in environmental genomics, offering valuable insights into the complex interplay between biological and environmental factors in oceanic ecosystems.
With regard to the second effort, we applied machine learning to generate the first spatio-temporal high-resolution model of organic matter abundance in the Ocean with the aim of describing the biological carbon pump at a global scale. A database has been consolidated, gathering satellite data and 8803 vertical profiles particulate organic matter gathered from 2008 to 2020 in different areas of the Ocean. We applied gradient boosting algorithms (XGBoost) combined with UMAP in order to create a global visualization. Special care has been taken to deal with the preparation of the ML training and the test data sets and the hyperparameter sweeps.We ran several experiments varying the input data, the spatio-temporal resolution of the working grid and the model architecture. A serial particle distribution with 10 particle size classes is predicted on a daily scale and a spatial resolution of 0.25 angular degrees, generating a surface particle field from the model, supporting an innovative description of the distribution of particulate organic matter in the Ocean.
In this context, P. Marquet and colleages has published a groundbreaking paper associating temperature and biological functions that is assimilated by different members of the team for in-depth study.
Models and physics-informed methods
There has been a substantial amount of work in the context of physics-informed ML.
A model based on neural ODEs was developed to monitor a phytoplankton biomass evolution inside an experimental photobioreactor. Two models were studied, the first one replaces the growth rate of the microalgae by a neural network, the second one uses a neural network to correct the bias of a kinetic model based on the literature. Mini batch training and learning rate schedulers were tested. We implement our own solver for ordinary differential equation problems, which interpolates the data and solves a set of equations in every step to allow mini-batch training. Both models efficiently reproduced the effect of a light gradient on a population of cell. The work will be submitted to Engineering Applications of Artificial Intelligence. Combining a host-virus model with different SST temperature predictions to represent long-term dynamics in environments revealed the potential for long-term host-virus coexistence, epidemic free or habitat loss states.
We generalized our model to variation in global sea surface temperatures corresponding to current and future seas and show that climate change may differentially influence virus-host dynamics depending on the virus-host pair. Temperature-dependent changes in the infectivity of virus particles may lead to shifts in virus-host habitats in warmer oceans, analogous to projected changes in the habitats of macro-, microorganisms and pathogens (work between Biocore and the LOV).
We addressed the coupling of hyperbolic PDEs with a neural network for the reconstruction of dust transport during the Holocene period. To our knowledge, this is the first time it has been applied in this field. This work has given rise to an article currently being submitted.
We approached the problem of PDE/NN coupling
in the thesis of L. Migus with a focus on the identification of a
friction term in a Saint-Venant-type system. The idea here is to make a network play the role of the term in a numerical solver, i.e., the network is called up at each time step, taking the flow height and velocity as input variables.
The difficulty lies in learning through a solver, since friction is not directly available; the only observable quantities are height and velocity. We study the resulting degradation and show that the resulting solver is however stable and highly efficient. This work is currently being written. Léon Migus’ thesis was defended in December 2023.
We worked on the coupling of Navier-Stokes and NPZ(D) models to create dynamic models of plankton populations in the ocean.
Finally, we proposed the Multi-Objective Physics-Informed Neural Networks (MOPINNs). MOPINNs applies evolutionary multi-objective AutoML to find the set of trade-offs between the data and physical losses of PINNs by at the same time optimizing the model complexity. It allows practitioners to correctly identify which of these trade-offs better represent the solution they want to reach.
This work has received the ``accessit to best paper award’’ at conference IBERAMIA’2022.
page not found
How it works
Oceania-query-fasta is a Python package that is built to make queries in Ocean Microbial Reference Gene Catalog v2 ~100GB (gziped) of FASTA, CSV and TSV files.
Install library oceania-query-fasta
Open complex Demo in Google Colab
See complex Demo in Jupyter NbViewer
Download simple Jupyter Notebook