tinyML Talks on January 23, 2024 “Advancing Medical Imaging Analysis with Multi-task and Hardware-Efficient Neural Architecture Search” by Hadjer Benmeziane from IBM Research Europe

We held our next tinyML Talks webcast. Hadjer Benmeziane from IBM Research presented Advancing Medical Imaging Analysis with Multi-task and Hardware-Efficient Neural Architecture Search on January 23, 2024.

The proliferation of electronic health records (EHR) has catalyzed a paradigm shift in healthcare, presenting an opportunity for leveraging artificial intelligence (AI) in medical data analysis. This talk introduces a novel benchmark in neural architecture search, expressly designed for optimizing AI models for edge deployment in EHR contexts. The benchmark synergizes multi-task learning with hardware-efficiency metrics, addressing the exigency of real-time, on-site decision-making in medical care. The discussion will elucidate the importance of applying HW-NAS to medical imaging, highlighting how it addresses the computational constraints and real-time processing requirements inherent in medical diagnostics. The creation process of this benchmark, which incorporates multi-task learning and hardware-efficiency metrics, will be detailed. Initial results demonstrating the benchmark’s impact in refining AI models for efficient and accurate medical image analysis will be presented, showcasing its potential to revolutionize healthcare diagnostics in resource-limited settings.

Dr. Hadjer Benmeziane is a visiting researcher at IBM Research Europe, specializing in hardware-aware neural architecture search for emerging AI accelerators such as analog in-memory computing. She received her PhD from Université Polytechnique des Hauts-de-France in August 2023, following her Master’s and Engineering degree in Computer Science from Ecole Supérieure d’Informatique, Algiers, Algeria. Her work on Analog Neural Architecture Search received the prestigious IEEE open source science award and best paper award at IEEE Services Computing 2023 Symposium. Her research focuses on making hardware-aware neural architecture search more efficient, flexible and practical.


Download presentation slides:
Hadjer Benmeziane

Watch on YouTube:
Hadjer Benmeziane

Feel free to ask your questions on this thread and keep the conversation going!