Federated Learning for Private, Fair, and EfficientAI for Well-being

Purpose and goal:

This project endeavors to enhance federated learning by incorporating self and semi-supervised learning methods. The primary goal of this research is to create efficient federated models with reduced or no supervision, addressing important AI challenges like data privacy, distribution, and labeling expenses. Furthermore, through the utilization of Domain Adaptation techniques in the federated setting, the project seeks to promote model fairness.

In summary, the project´s aim is to develop efficient, private, and fair models suitable for in-home monitoring applications. Expected results and effects:

This project addresses real-world challenges in in-home monitoring via semi/self-supervised federated learning. It tackles issues of reducing labeling costs and data privacy, which are crucial when handling personal data. By utilizing Domain adaptation, it ensures gender fairness.

So, the results will be around efficiency, privacy, and fairness, aligning with society´s needs. This project will be done by Zahra at CAISR lab in Sweden and will serve as the foundation for long-term collaboration with Queen´s University´s Aiim lab in Canada. Approach and implementation:

The project will be done in a total of 18 weeks in collaboration with Aiim lab. Zahra will lead and execute the project with mentorship from Prof, Ali Etemad. The planned visit schedule was as follows: First visit: 10 weeks (Oct 16th to Dec 23rd, 2023). Second visit: 8 weeks (April 1st to May 25th, 2024).

As mentioned in the proposal, Zahra requires a visa to travel to Canada. Currently, Zahra is awaiting her visa, which has been delayed, subsequently affecting the project´s timeline. However, the duration of the delay will be accommodated within the time gap between the two scheduled visits.