Print

Print


ECE Department Seminar 

 

Machine Learning at the Edge  

Algorithmic Design Considerations for Healthcare Applications 

 

Miad Faezipour, Ph.D. 

Associate Professor 

Computer Science and Engineering and Biomedical Engineering 

University of Bridgeport 

 

Friday, February 26, 2021 

10:00 am 11:00 am 

Zoom Meeting Link: 

https://gmu.zoom.us/j/98236145469 

 

 

Abstract: Recent trends in clinical and telemedicine applications highly demand automation in the processing, analysis, and augmentation of the biomedical data for designing intelligent decision support systems as well as for high reliability and interpretability of their machine learning-based inferences. Due to the security and privacy issues inherent to the medical data, cloud computing resources may not be a viable solution in many diagnostic or therapeutic scenarios. Therefore, there is a pressing need for on-device data analysis in evolving medical and healthcare applications. Moreover, for many critical/urgent cases, medical decisions based on data processing must be made immediately, while cloud connectivity may not be available or reliable. My research is centered around designing advanced machine learning and signal processing algorithms that are efficiently implementable in-situ, from both data constraint and time complexity perspectives. I also address the medical interpretability requirement by integrating domain expert knowledge into the algorithmic design pipeline. My talk today spans various D-BEST (Digital/Biomedical Embedded Systems and Technology) lab research areas, mainly covering 3 emerging healthcare applications, with underlying bases for machine learning based edge-computing. First, a novel cardiac behavior profiling scheme is introduced to classify irregular heartbeats from normal ones, with special considerations for digital hardware implementations. Second, the analysis of the acoustic signal of respiration is discussed in order to augment a virtual reality therapy platform for assisting breathing regulation in an embedded device. Third, computational models of the basal ganglia neurons are discussed with applications in neuromorphic hardware design and adaptive parameter selection for deep brain stimulation. These research efforts lie at the intersection of signal/image processing, computer vision and deep learning with digital/embedded software/hardware co-designs. The premises ideas call for further data dimensionality augmentation and interpretable representation learning algorithms in embedded devices using the integration of domain knowledge with advanced machine learning techniques.  

 

Bio: Miad Faezipour, PhD is an Associate Professor of Computer Science and Engineering and Biomedical Engineering at the University of Bridgeport (UB), CT and the founder and director of the Digital/Biomedical Embedded Systems and Technology (D-BEST) research laboratory since 2011. Prior to joining UB, she has been a post-doctoral research associate at the University of Texas at Dallas collaborating with the Center for Integrated Circuits & Systems (CICS) and the Quality-of-Life Technology (QoLT) laboratories. She received the B.Sc. in electrical engineering from the University of Tehran, Iran and the M.Sc. and Ph.D. in electrical engineering from the University of Texas at Dallas. Her research interests and expertise lie in the areas of digital/biomedical embedded hardware/software co-designs, biomedical signal/image processing, computational neuroscience, computer vision and healthcare/biomedical informatics, as well as machine/deep learning, artificial intelligence (AI) and AI-based bio-data augmentation. She is a senior member of IEEE, EMBS and IEEE Women in Engineering.