Research Intern
I worked on the project titled "Design and implementation of an energy-efficient hardware accelerator for LSTM-based federated learning on FPGA platforms" under the supervision of Dr. Shervin Vakili at ECCoLe Lab, University INRS. In this project we introduced an accuracy-aware approximate hardware core designed for the efficient computation of the nonlinear functions involved in on-device training of LSTMs. Our approach divides the functions into four regions based on their curvature, applying the most appropriate approximation technique to each region. We optimized the parameters of our model to achieve an optimal trade-off between hardware cost and accuracy, demonstrating superior performance compared to existing approximation methods.
Jun 2024 - Sep 2024
