About the project
-
Value
$1 752 189
-
Prompt Contribution
$739 291
Training convolutional neural networks (CNNs) for assisted/autonomous driving requires a massive amount of data. The training data comes from multiple geographical areas, each of which needs to be represented in order to achieve robust training.
Each type of object to be detected and identified (signs, pedestrians, etc.) requires a CNN.
Geographical morphological variations require re-training for each zone. Keeping the trainings up to date also adds to the problem, forcing the approach we take. So here we see a problem in terms of the amount of data and the processing time that will be required.
Project objective
The aim is to speed up CNN training and keep training local to the capture site, by combining QML and AF. AF will reduce the burden on compute nodes by using multiple servers located in the capture region. QML will map a continuous progress path that will accelerate the training process on each of these compute nodes.
Impacts
Leddartech will increase the number of highly qualified French-speaking quantum computing personnel. This project will accelerate the transfer of knowledge from the Quantum Institute to Leddartech, while using a 100% Québec computing platform, PINQ2. AF promises to cover a field of application hitherto unseen in our field. Ultimately, QML combined with FA should enable continuous training of the world’s road network. The quantum algorithm research we will be carrying out will enable us to identify the next steps in mapping and achieving real-time inference, another significant and profitable benefit for Leddartech.
Organisation (s)
