General information
Organisation
The French Alternative Energies and Atomic Energy Commission (CEA) is a key player in research, development and innovation in four main areas :
• defence and security,
• nuclear energy (fission and fusion),
• technological research for industry,
• fundamental research in the physical sciences and life sciences.
Drawing on its widely acknowledged expertise, and thanks to its 16000 technicians, engineers, researchers and staff, the CEA actively participates in collaborative projects with a large number of academic and industrial partners.
The CEA is established in ten centers spread throughout France
Reference
2025-36978
Description de l'unité
The Embedded Artificial Intelligence Laboratory (LIAE) is responsible for designing, developing and implementing solutions based on neural networks optimised for embedded systems.
By joining our dynamic and experienced team, you will contribute to the development of our expertise in new neural networks for embedded systems, which are highly sought after in the French and European industry of tomorrow.
Position description
Category
Engineering science
Contract
Internship
Job title
Neural approximation of non-linear operations for efficient inference of new emergent model
Subject
The enthusiasm for new neural network models is justified by their very high performance compared to existing CNN models. Indeed, they outperform the latter in a large number of application tasks. However, they are relatively complex and their deployment in the embedded domain faces several challenges. Among other things, non-linear functions are ubiquitous in these models and often require the use of complex operations manipulating floating point data. Therefore, as part of its activities in lightweight inference design for embedded systems, LIAE is interested in implementing non-linear operations by constructing AI-based approximations.
Contract duration (months)
6 months
Job description
The aim of the internship is to consider a complete implementation while getting as close as possible to the performance of the reference network. The implementation of quantization methods may also be addressed in a related manner.
In this context, the objective of this internship is to identify the non-linear functions involved in ViT/VisionMamba networks and to implement learning techniques for generating approximate non-linear operations in an embedded context. The candidate's main tasks will be as follows:
· Identifying the non-linear operations involved,
· Deploying and validating the learning approximation method through simple tests,
· Integrating the operators into ViT and Vision Mamba models on reference and quantified models.
Methods / Means
Linux/Python/Pytorch/ML
Applicant Profile
You are preparing an Engineering degree in embedded systems and signal processing
You are passionate about scientific and technological researches. Ideally, you have some initial experience working with the Pytorch environment in the context of implementing neural networks. Knowledge of hardware architectures is a plus. Proficiency in reading scientific reference papers in English would also be appreciated.
Position location
Site
Saclay
Job location
France, Ile-de-France, Essonne (91)
Location
Palaiseau
Candidate criteria
Languages
English (Intermediate)
Prepared diploma
Bac+5 - Diplôme École d'ingénieurs
Recommended training
Systèmes embarqués; Traitement de signal
PhD opportunity
Non
Requester
Position start date
05/01/2026
Create candidate area
