Thèse CIFRE / PhD - Deep Learning / Graph Neural Networks

Led by Rodolphe Saadé, the CMA CGM Group, a global leader in shipping and logistics, serves more than 420 ports around the world on five continents. With its subsidiary CEVA Logistics, a world leader in logistics, and its air freight division CMA CGM AIR CARGO, the CMA CGM Group is continually innovating to offer its customers a complete and increasingly efficient range of new shipping, land, air and logistics solutions.

Committed to the energy transition in shipping, and a pioneer in the use of alternative fuels, the CMA CGM Group has set a target to become Net Zero Carbon by 2050.
Through the CMA CGM Foundation, the Group acts in humanitarian crises that require an emergency response by mobilizing the Group’s shipping and logistics expertise to bring humanitarian supplies around the world.

Present in 160 countries through its network of more than 400 offices and 750 warehouses, the Group employs more than 155,000 people worldwide, including 4,000 in Marseilles where its head office is located.

Collaborating Academic Institution
 

The academic part of this PhD position will be co-supervised by Prof. Johannes Lutzeyer and Prof. Michalis Vazirgiannis at École Polytechnique. École Polytechnique is the premier engineering University of France and a founding member of the recently established Institut Polytechnique de Paris (which entered the international rankings in high positions). Famous scientists (including Nobel prize recipients) and industrial leaders are alumni of the school, offering an exceptional environment for research in the fast-growing excellence pole of Saclay, hosting a rich ecosystem of industrial and academic research centers a few kilometers south of Paris. The Data Science and Mining group, in which you would also be integrated, has already had significant impact in local and international research and industrial activities with several high-impact publications and successful industrial projects.

PhD Position and Context:

CMA CGM and École Polytechnique open their first joint PhD position on the subject of “Graph Neural Networks to predict and simulate properties and structure of the evolving maritime transportation network”. As part of this project, we are seeking a PhD researcher with an excellent background in machine learning and deep learning.

The PhD candidate will work closely with the IT Data Science team of CMA CGM as well as with the Computer Science laboratory of École Polytechnique. The aim of the thesis is to develop new graph neural network approaches applied to maritime transport data and other external data sources. The student will have access to all internal tools (Dataiku, eks, GPUs...) and a rich scientific environment covering all aspects of AI, including many seminars, workshops and gatherings for PhD candidates and postdocs.

Graph neural networks are promising candidates to accurately predict the evolution of the structure and different quantities on the maritime transportation network. The proposed research falls into the extremely active area of temporal graph neural networks that is currently attracting a lot of academic and industrial attention with a common set of reference benchmarks being published only last year [1] (with a recent extension earlier this year [2]). Coupling this exciting area of progress with the wealth of relevant data available at CMA CGM is a great environment for innovative progress in this scientific area both from an academic and industrial perspective.

The 3-year thesis will be located in Marseille at the CMA CGM Office for the first two years with monthly trips to the Saclay plateau. The final year will take place at Ecole Polytechnique.

[1] Huang et al., "Temporal graph benchmark for machine learning on temporal graphs," NeurIPS, 2023.

[2] Gastinger et al., "TGB 2.0: A Benchmark for Learning on Temporal Knowledge Graphs and Heterogeneous Graphs," arXiv:2406.0963, 2024.

PhD Student Responsibilities:

The goal of this PhD is to develop innovative temporal graph neural networks using data related to maritime logistics and other relevant features to predict the evolution of the structure and features of the global containerized shipping network. A preliminary study has been conducted which highlights both slow and abrupt evolutions of the concerned network. These evolutions have significant impact (on e.g., the environment, productivity, congestion, etc.). The aim is therefore to predict these evolutions in their multidimensional components and examine the relationship between the graph structure and the economy (freight rates).

Specific objectives are as follows:

enrich the existing pipeline of CMA CGM with more advanced deep learning models, more tasks and more datasets
gain proficiency in operational maritime data
propose adequate and novel graph-based methodology including temporal graph neural networks
perform benchmarking experiments across deep learning (and also standard machine learning) models on real and simulated datasets to create a new standard for the community
illustrate the evolutions of the global containerized shipping network and characterize its impact
implement the approaches/methodologies in Dataiku and share the knowledge to the other Data Scientists of the company and research lab.
participate at national and international conferences and publish scientific articles

Requirements:

Excellent background in theoretical and empirical aspects of machine learning and deep learning
Good background in Statistics and Linear Algebra
Excellent programming skills in Python
Good writing skills as well as relational and communication skills with a strong capacity to collaborate effectively with other team members (multicultural and international environment)
A proficient level in English

Additional Desired Qualifications:

Knowledge of the specificities of graph neural networks
First experiences with logistics data
Good programming skills in SQL

Come along on CMA CGM’s adventure !