查看文章

Currently, machine learning techniques are being extensively applied in areas such as natural language processing, general pattern recognition, object detection in images, visual place recognition etc. The advent of Deep Neural Networks has given the computers image recognition capabilities that are now at par with humans. However, this capability comes at the price of extremely high computational demand which is generally out of reach for conventional CPUs. Thus, GPUs have been considered for applications which require real-time processing. GPUs can handle vector computations at higher speeds and can thus process convolution operations quickly. Moreover, deep networks require extensive time and enormous data for training. However, transfer learning approaches make it feasible to consider retraining for custom tasks relatively quickly, and with fewer training examples and also take care of the over …