Machine Learning Systems: Low-Energy VLSI Architectures and Applications

Machine learning and data analytics continue to expand the fourth industrial revolution and affect many aspects of our lives. This talk will explore machine learning applications in datadriven neuroscience, and low-energy implementations of machine learning systems. Datadriven neuroscience can exploit machine learning approaches including deep learning to generate hypotheses associated with biomarkers for specific neuro-psychiatric disorders. In the first part, I will talk about use of machine learning to find biomarkers for epilepsy and adolescent mental disorders such as borderline personality disorder (BPD), using electroencephalogram (EEG) and functional magnetic resonance imaging (fMRI), respectively. In the second part of the talk, I will talk about approaches for energy-efficient implementations for both traditional 2019-2021 CASS Distinguished Lecturer Roster machine learning and deep learning systems. I will talk about the roles of feature ranking and incremental-precision approaches to reduce energy consumption of traditional machine learning systems. I will then talk about our recent work on Perm-DNN based on permuted-diagonal interconnections in deep convolutional neural networks and how structured sparsity can reduce energy consumption associated with memory access in these systems.