Few-Shot Learning on Graph-Structured Data: the Case of Brain Activation Maps
Myriam Bontonou (PhD candidate, IMT Atlantique, Lab-STICC)
Title : Few-Shot Learning on Graph-Structured Data: the Case of Brain Activation Maps
Asbtract : Deep learning is state-of-the-art in many fields as long as a large amount of data is available. Yet, sometimes, this condition is not met. This is why, in recent years, new deep learning methods have been developed to solve problems with few training examples.
In this presentation, we will first show that a major question still arises about the generalization ability of few-shot learning methods. Then, we will address the particular case of neuroimaging data. We will show how few-shot learning methods can be applied to these complex data, although there is still progress to be made to fully exploit their structure.