Within the Institute and the Department of Ambient Intelligence and Interactive Systems (DIASI), and the Artificial Intelligence, Language, and Vision Service (SIALV), the Laboratory for Semantic Analysis of Text and Image (LASTI) conducts foundational research projects in artificial intelligence applied to computer vision, automated processing, and multimodal data analysis.

The global acceleration of AI research has made the field extremely competitive. To maintain its scientific excellence and technological competitiveness, LASTI—in partnership with other laboratories within the service—aims to expand its capacity for developing and operationally implementing AI-based systems. The design process for AI models requires a high level of proficiency in software environments that facilitate the handling of data, annotations, and models, as well as their integration into complex systems across various computational scales (ranging from the training and evaluation of models of varying sizes to multi-agent AI systems and AI models deployed on physical systems).

As an intern at the CEA, you will have the opportunity to work within a world-renowned research environment. Our teams consist of passionate and dedicated experts, providing an ideal setting for learning and collaboration. You will have access to state-of-the-art equipment and top-tier research resources to carry out your assignments. The work conducted may potentially lead to a scientific publication.

Context :

Most image-recognition systems are trained with labeled examples and a fixed list of categories. After deployment, however, new images continue to arrive. Their visual appearance may change because of a different style, camera, or environment, and the system may also encounter categories absent from its initial training data. In continual category discovery a model first receives labeled images from known categories, then successive collections of unlabeled images containing both known and genuinely new categories. It must recognize known categories, form groups for new ones, and retain what it has learned over time. The main challenge is that visual similarity does not always mean that two images show the same category. A known vehicle represented as a sketch may look less similar to photographs of the same vehicle than to a visually close but new vehicle category. The internship will investigate methods to resolve these ambiguities with limited expert feedback.

What do we expect from you ?

The intern will work on the following tasks :
Conduct a literature review on methods for category discovery, visual domain adaptation and active learning
Analyze when changed appearances of known categories are confused with genuinely new categories
Select informative images for expert feedback under a limited query budget
Use this feedback to improve category discovery and maintain categories over time
Evaluate the approach against fully automatic and randomly queried baselines

This internship offers an introduction to research in machine learning and computer vision. Depending on the results, it may lead to a scientific publication and may be pursued with a PhD focused on a similar topic.

#Cea-List

• Students in their final year of studies (M2 or last year of engineering school)
• Strong foundations in machine learning, deep learning and computer vision
• Interest in visual representation learning, continual discovery, or active learning
• Python proficiency in a deep learning framework