5+ Techniques for Efficient Implementation of Neural Networks

Embedding real-time, large-scale deep learning vision applications at the edge is challenging due to their huge computational, memory and bandwidth requirements. System architects can mitigate these demands by modifying deep neural networks (DNNs) to make them more energy-efficient and less demanding of embedded processing hardware. In this talk we’ll provide an introduction to today’s established techniques for efficient implementation of DNNs: advanced quantization, network decomposition, weight pruning and sharing and sparsity-based compression. We’ll also preview up-and-coming techniques such as trained quantization and correlation-based compression.