The Conglomerate Compression Approach for Visual Pattern Recognition Networks in Edge AI Applications

Deep Neural Networks have shown remarkable performance in various applications of Visual Pattern Recognition (VPR). This field continues to grow due to emergence of new architectures, availability of huge data, and powerful computing resources. However, because of their high computational complexity that subsequently demands higher memory requirements, and power consumption, the deployment of such DNN-based Visual Pattern Recognition Networks (VPRN) models in edgeAI applications is hindered. Such overheads necessitate the compression of VPRN models while retaining their performance. This paper explores multiple compression techniques which combines Knowledge Distillation (KD), pruning and quantization that are applied to DNNs to achieve higher compression ratio.