Title. Explaining the trajectories of jellyfishs with heterogeneous text and images mining and visualization Key words. Ecology, Deep learning, Explainability, Visual Analytics, Analysis of spatiotemporal series, Heterogeneous data visualization. Context. The study of jellyfish outbreaks is important because they can have a negative impact on the provisioning of marine ecosystems services for human welfare. In addition, they can disrupt human activities, such as fishing, tourism and shipping. Jellyfish outbreaks are often linked to environmental changes such as water temperature, salinity and currents. Studying these factors can help predict future jellyfish outbreaks and develop management strategies to mitigate their impact on the environment and human activities. There are several methods to collect data on jellyfish, to better understand their distribution, abundance and behavior such as visual observations- as jellyfish can be visible on the surface of the water-, the use of plankton nets, satellite tags to track their movements, underwater cameras, etc. However, their capture and conservation and their fleeting and unpredictability appearances induce scarce scientific datasets. Objectives. We here propose the collection and analysis of multilingual and heterogeneous documents to complement existing data (i.e., JeDI) with original data social media, press and scientific literature. Our objective is to depict patterns and trends in collected data, to map spacetime patterns of coastal aggregations and stranded jellyfish, and to track their diversity changes and the contribution of non-indigenous species to such phenomena (e.g. determination of invasion speed). The originality of our project is to disentangle the extent to which these events and the distribution patterns of non-indigenous species are related to climatic change and ecosystem degradation. Methodology. We will follow a three steps strategy: 1) analysis of spatio-temporal series: multimodal supervised classification to exploit heterogeneous resources and spatio-temporal clustering to group similar data in space and time and to use species distribution modelling to understand their relation to the environment and model future evolutions 2) heterogeneous data visualization: design, implementation and validation of an interactive visual interface for exploring the results of the first step (classes, patterns, clusters…) and the texts/images from the raw data. Particular attention will be paid to the 3) interpretability of the used methods. Deadline for applications: 30 May 2023 Duration of the PhD: 3 years Requirements: - a master's degree in Computer Science with successful research experience - advanced programming skills (design and implementation) - a good academic level attesting to his/her ability to combine practice and theory - a level of professional oral and written English - general knowledge in the field of artificial intelligence - an appetite for ecological issues Procedure and contact - Send tosandra.bringay@lirmm.fr,arnaud.sallaberry@lirmm.fr,maximilien.servajean@lirmm.fr - Your master's degree (if already obtained) and your transcripts - Curriculum vitae with 2 references - At least one letter of recommendation 
- Any publications you may have Applications are managed on a case-by-case basis. You will be informed promptly by email of the admissibility of your application and if you are invited to a first interview. We look forward to your applications! Sandra Bringay, Arnaud Sallaberry, Maximilien Servajean