An Integrative Algorithm/Architecture Co-Design Of Deep Spatial and Temporal Separable Convolutional Neural Networks

This dissertation presents my research on algorithm/architecture co-design of deep spatial and temporal separable convolutional neural networks and their applications. I will introduce DeepDive as a framework for enabling and power-efficient execution of spatial deep learning models on embedded FPGA. For emerging Deep Separable Convolutional Neural Networks (DSCNNs), DeepDive is a fully-functional, vertical co-design framework for power-efficient implementation of DSCNNs on edge FPGAs. Agile Temporal Convolutional Network (ATCN) is also proposed for high-accurate fast classification and time series prediction in resource-constrained embedded systems. ATCN is primarily designed for mobile embedded systems with performance and memory constraints, such as wearable biomedical devices and real-time reliability monitoring systems. It uses the separable depth-wise convolution to reduce the computational complexity of the model and residual connections as time attention machines, to increase the network depth and accuracy. The result of this configurability makes the ATCN a family of compact networks with formalized hyper-parameters that allow the model architecture to be configurable and adjusted based on the application requirements. I also will present DeepTrack and DeepRACE, which are two other aspects of the application of DNN in vehicle trajectory prediction in highways and real-time reliability monitoring of transistors.