This project will develop algorithms and tools to provide a generic FL framework to deploy trustworthy and energy-efficient ML applications onto heterogeneous edge devices. It uses a cross-layer approach to utilize the full potential of the FL processing flow from communication scheduling to hardware deployment. To this aim, we will (i) develop adaptive scheduling mechanisms that guarantee the system predictability; (ii) design scalable and energy-efficient heterogeneous machine learning (HML) models customized for each FL node individually with guaranteeing user requirements; (iii) design trustworthy FLS that are resistant to different types of cyber-attacks as well as noisy environments; and (iv) facilitate fast deployment and effective maintenance of HML models.
Enhancing Global Model Performance in Federated Learning with Non-IID Data using a Data-Free Generative Diffusion Model (Oct 2024) MohammadReza Najafi , Masoud Daneshtalab, Jeong-A Lee , Seokjoo Shin Journal of IEEE Access (IEEE-Access)
A Systematic Literature Review on Hardware Reliability Assessment Methods for Deep Neural Networks (Jan 2024) Mohammad Ahmadilivani , Mahdi Taheri , Jaan Raik , Masoud Daneshtalab, Maksim Jenihhin ACM Computing Surveys (CSUR)
Exploration of Activation Fault Reliability in Quantized Systolic Array-Based DNN Accelerators Mahdi Taheri , Natalia Cherezova , Ali Mahani , Maksim Jenihhin , Masoud Daneshtalab, Jaan Raik International Symposium on Quality Electronic Design (ISQED'24)