Missions

The person hired for this position will work within the context of the French Océan et Climat program, particularly in the AI data challenges part of the project
https://www.ocean-climat.fr/Le-PPR/Actualites/Intelligence-artificielle-Appel-a-postdoc-et-data-challenges
The goal is to have several teams "coopeting" (collaborating and competing at the same time) to solve major data challenges in marine science. The challenge of interest here is the prediction of the composition of communities of marine entities (zooplankton, marine snow particles, and coral reef fishes) from environmental data taken around the point of the biological observations. Formally, this means that the output is multivariate (from 10 to 600 groups) and that the input has up to four dimensions (latitude, longitude, depth, time) for each variable (and there are over a dozen input variables). The foreseen approach would be to build on recent developments in using CNNs for Species Distribution Models (e.g. Deneu et al 2021, Morand et al 2024) to summarise the complex input fields and then regress on the target variables in the embedding space. However, many challenges remain to be solved with this approach (regarding the use of CNNs vs Vision transformers, the structuring of the input, the inclusion of depth and time dimensions, the loss function for the multivariate output, etc.) and we therefore expect the post-doc to make significant contributions to the field over the course of the project.

References:
- Deneu B et al (2021) Convolutional neural networks improve species distribution modelling by capturing the spatial structure of the environment. PLoS Comput Biol
- Morand G et al (2024) Predicting species distributions in the open ocean with convolutional neural networks. Peer Community J

Activités

Main activities (in chronological order):
- Familiarise oneself with the various input and output datasets
- Investigate the loss function for a multivariate regression output with very different ranges in the target variables and unknown density distributions
- Investigate the best way to summarise the 2D inputs (longitude, latitude) to capture the context of the observation and improve the prediction
- Investigate the addition of a third and fourth dimensions to the inputs

Other activities : The post-doc is expected to collaborate with the other team working on the same challenge, based in Montpellier. He/she will also be involved in general activities related to the Océan and Climat program as well as other ongoing projects involving the supervisors, such as the OceanIA team of INRIA https://oceania.inria.cl/. Finally, he/she will also work closely with a PhD student currently investigating a similar approach to predict phytoplankton community composition.

Compétences

Transversal knowledge required :
- Expertise in machine learning and deep learning in particular
- Knowledge in ecology, marine biology, or oceanography would be a plus
- Good writing skills and oral expression in english (at least B2)
- Scientific rigor and curiosity

Technical skills :
- Required: PhD or engineering degree in deep learning
- Required: good programming skills and ability to deal with large datasets
- Required: publication of at least one research article in a scientific journal or major conference as first author
- Optional: previous experience in ecology, marine biology, or oceanography
- Optional: previous experience with species distribution models

People skills :
- Both autonomy and teamwork abilities
- Collaboration in an international context

Contexte de travail

The Laboratoire d'Océanographie de Villefranche (LOV ; http://lov.obs-vlfr.fr/) is located close to Nice, on the French Riviera. It belongs to one of the three marine stations of Sorbonne Université. With about 120 permanent staff, the LOV generates and analyses a large quantity of marine data, including imaging, genomic, and satellite data to study the ocean.

The COMPLEx (COMPutational PLAnkton Ecology) team gathers about fifty members studying marine plankton by collecting data with quantitative imaging instruments and high throughput genomics that informs advanced numerical analysis methods (modeling, statistics, machine learning). Plankton encompasses all organisms roaming with marine currents. Those organisms are responsible for producing some of the oxygen we breathe, storing the carbon we emit, feeding the fish we eat; plankton is therefore a major building block of Earth's ecosystem. COMPLEx strongly interacts with the Quantitative Imaging Platform of Villefranche (PIQv; https://sites.google.com/view/piqv), which oversees the operation of the tools that the team develops. Those tools include imaging sensors, such as the Underwater Vision Profiler or the ZooScan, as well as an increase number of software packages to process and control the quality of the data generated by the instruments, sort images taxonomically (https://ecotaxa.obs-vlfr.fr/) or store and distribute data on the abundance of marine snow particles. The team has a long experience of interactions with engineers and computer scientists, in academia and the private sector, to develop these tools.
The post-doc will be directly supervised by Jean-Olivier Irisson https://www.obs-vlfr.fr/~irisson/. Prof Irisson is a computational ecologist, working at the interface between plankton ecology and computer sciences.

The postdoc will work in strong connection with the GreenOwl team of Inria (National Institute for Research in Digital Science and Technology). The first aim of Greenowl is to assess the resilience capability of microbial ecosystems to climate change. The second aim is taming the adaptation of microorganisms, to develop new energy or protein sources, new ways to recycle carbon, nitrogen and phosphorus. Greenowl deploys theoretical tools, from the fields of automatic control and artificial intelligence and combines them with experimental validation on its experimental platforms.
The post-doc will be co-supervised by Olivier Bernard (https://www-sop.inria.fr/members/Olivier.Bernard/). Dr Bernard is a specialist of phytoplankton modelling and theoretical approaches of the ocean system.

Le poste se situe dans un secteur relevant de la protection du potentiel scientifique et technique (PPST), et nécessite donc, conformément à la réglementation, que votre arrivée soit autorisée par l'autorité compétente du MESR.

Contraintes et risques

None

Informations complémentaires

Paid vacation up to 55 days per year.
The laboratory is located in Villefranche-sur-mer and has direct access to the sea.