Annonce

 

Duration: 6 months

 

Supervisors: Nisrine Jrad, Patty Coupeau, Jean-Baptiste Fasquel and Fahed Abdallah

 

Key-words: Deep learning; Graph neural networks (GNNs); Social Networks; Transport Networks; Computer Vision, Physical inspired graph

 

Title: Graph Neural Networks for Social Networks

 

Context: In the last decade, Deep Learning, particularly Convolutional Neural Networks (CNNs), has significantly influenced several machine learning applications, including image recognition [1] and signal processing [2]. These accomplishments have primarily involved sequences (1D) or images (2D) data structured on grid formats that leverage linear algebra operations in Euclidean spaces. However, numerous domains, such as social networks, molecules, and knowledge graphs, pose challenges as their data cannot be straightforwardly encoded into Euclidean domain but lends itself to natural representation as graphs.

Graphs serve as a robust tool for representing data generated by both artificial and natural processes. They possess a dual nature, being composed of atomic information pieces or entities (known as nodes) while exhibiting relational structures defined by links among entities (edges). The omnipresence of graphs is noteworthy. In social sciences, graphs are used for depicting interpersonal relationships, in recommending systems they model intricate purchasing patterns. In social networks they are used to identify communities, and in traffic networks, they can predict traffic and optimise routing.

The increasing wealth of graph data, along with the expanding accessibility of extensive repositories, has encouraged researchers to extend the deep learning paradigm to graph world. The objective is to revisit Neural Networks to operate on graph data, in order to benefit from the representation learning ability. In this context, many Graph Neural Networks (GNNs) have been recently proposed in the literature of geometric learning.

In this work, we will focus on node classification task with model-guided GNN. More specifically, we will be interested to integrate “a model graph”, inspired by a domain-specific knowledge, together with graph data in the GNN learning process. The incorporation of these expert knowledge into the learning of GNNs holds the potential to constrain and streamline their training process, thereby improving overall generalization and accuracy. Our research team has explored the use of GNNs as an effective tool of semantic segmentation [3,4]. They also proposed a first approach to design a model-guided GNNs and applied it to image segmentation.

 

The main objective of this research is to propose a guided model GNNs and apply it to Social Networks data. This will be done through several steps:

·Conduct a literature review on Graph Neural Networks (GNNs).

·Develop a physical inspired GNN for social networks applications.

·Apply it to an open-access dataset.

·Compare the performance with state-of-the-art approaches.

·Present the results in a collaborative meeting.

·Publish the open-source code and the reproducible analysis pipeline.

·Write a report or a scientific article on the obtained results.

 

Research environment/Location: The research will take place within the LARIS laboratory in Angers, France. The internship will be supervised by Nisrine Jrad, Patty Coupeau, Jean-Baptiste Fasquel and Fahed Abdallah

 

Candidate profile: Candidates are expected to be graduated in computer science and/or machine learning and/or signal image processing and/or applied mathematics/statistics. The internship subject requires skills in Python development tools, specially TensorFlow or Pytorch.

 

Salary: According to the internship gratification scale (600 euros per month approximately)

 

For more details: Feel free to contact supervisors

 

To apply: please send a CV, a cover letter, and Master 1 and Master 2 grades, and email address of two referees to N. Jrad (njrad@uco.fr), J.B Fasquel (Jean-Baptiste.Fasquel@univ-angers.fr) and F. Abdallah (fahed.abdallah76@gmail.com)

 

[1] Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 25.

[2] Abdallah, T., Jrad, N., Abdallah, F., Humeau-Heurtier, A., & Van Bogaert, P. (2023). A self-attention model for cross-subject seizure detection. Computers in Biology and Medicine, 165, 107427.

[3] Chopin, J., Fasquel, J. B., Mouchère, H., Dahyot, R., & Bloch, I. (2023). Model-based inexact graph matching on top of DNNs for semantic scene understanding. Computer Vision and Image Understanding, 103744.
[4] Coupeau, P., Fasquel, J. B., & Dinomais, M. On the Use of Gnn-Based Structural Information to Improve Cnn-Based Semantic Image Segmentation. Available at SSRN 4214110.