IMT Atlantique, internationally recognized for the quality of its research, is a leading general engineering school under the authority of the Ministry of Industry and Digital Technologies, ranked in the 3 main international rankings (THE, SHANGAI, QS). With three campuses in Brest, Nantes and Rennes, IMT Atlantique aims to combine digital technology and energy to transform society and industry through training, research and innovation.
The successful candidate will work in the Computer Science Department, within the Maths&Net research team. The Maths&Net team, of the Cyber & Networks (CYR) pole of the UMR LabSTICC, aims at designing, describing, managing, securing, and controlling various aspects of operator communication networks, as well as other types of networks such as social networks. In a layered model we are at the network layer (IP) and above.
The successful candidate will also work closely with the MEE department (Mathematical and Electrical Engineering, Brest, France).
Machine learning models have demonstrated their capabilities of solving complex tasks, as long as enough data is provided. Realistic data, in communication networks, are hard to obtain or even hard to generate. Therefore researchers and industrials are using scaled down network simulators to generate a huge amount of data, expecting that trained machine learning model on such datasets can be naturally transferred on realistic datasets. Instead of asking ourselves which machine learning model has good transfer properties, we focus more on a data-centric approach: we try to identify the dataset on which a machine learning model should be trained to maximize its transferability property to a realistic dataset. In this project, we will study the transferability of machine learning models for communication network data. The goal is to provide a theoretical framework to answer this challenge and subsequently derive efficient algorithms to solve this problem on real datasets.
Aim 1 : Establishing a rigorous mathematical framework to quantify the transferability of DNN.
Aim 2 : Formulate and efficiently solve the data selection meta-learning problem with performance guarantees.
Aim 3 : Apply the methods obtained on realistic datasets
PhD less than 3 years old in the field of applied mathematics or computer science