Towards Fast GPU-based Sparse DNN Inference: A Hybrid Compute Model

An End-to-End Programming Model for AI Engine Architectures

“…• Edge Devices: Recently edge platforms (mobile phones, wearables, IoT sensors) have also become feasible deployment targets for DNNs, owing to advances in research on DNN architectures that maintain accuracy while reducing resource consumption (such as quantized (Wu et al, 2016) and sparse networks (Xu et al, 2018)). These advances notwithstanding, those platforms reproduce many of the phenomena of their larger scale analogues (Suo et al, 2021).…”