Creating and sharing knowledge for telecommunications

Project: Precise and Efficient Learning using Attention Mechanisms

Acronym: PRELUNA
Main Objective:
In this project, the objective is to apply attention mechanisms to improve the performance of neural networks in computer vision tasks. There is ample evidence that attention mechanisms can be used to improve the performance in many machine learning tasks, including medical image analysis, but these attention mechanisms are added to the convolutional neural networks without a principled approach. The success of the project will be measured by how much we advance the state of the art in specific medical image tasks, which will serve as use cases. The two use cases that will be used to assess the effectiveness of the approach are the diagnosis of ischemic stroke severity from brain computed tomographies (brain-CT) and the diagnosis of coronary artery stenosis, from X-ray angiographies.
Our general strategy is to use attention to focus the networks in the regions of interest, which will improve the classification accuracy of the systems. In the process, we may also shed light on the important relations between the phenomena of attention, consciousness, and human ability to learn from small samples.
Reference: PTDC/CCI-INF/4703/2021
Funding: FCT
Start Date: 01-01-2022
End Date: 31-12-2024
Team: Mario Alexandre Teles de Figueiredo, André Filipe Torres Martins
Groups: Pattern and Image Analysis – Lx
Partners: INESC-ID
Local Coordinator: Mario Alexandre Teles de Figueiredo

Associated Publications