CNN inference acceleration on limited resources FPGA platforms_epilepsy detection case study

<span lang="EN-US">The use of a convolutional neural network (CNN) to analyze and classify electroencephalogram (EEG) signals has recently attracted the interest researchers identify epileptic seizures. This success come with an enormous increase in computational complexity memory requirements CNNs. For sake boosting performance CNN inference, several hardware accelerators have been proposed. The high flexibility field programmable gate array (FPGA) make it efficient accelerator for Nevertheless, resource-limited platforms, deployment models poses significant challenges. ease implementation on such tools frameworks made available by research community along different optimization techniques. In this paper, we proposed FPGA automatic seizure detection approach using two models, namely VGG-16 ResNet-50. To reduce model size computation cost, exploited approaches: pruning quantization. Furthermore, presented results discussed advantages limitations alternatives inference acceleration quantized CNNs Zynq-7000: advanced RISC machine (ARM) software implementation-based ARM, NN, development kit (SDK) software/hardware deep learning processor unit (DPU) DNNDK toolkit.</span>