ABSTRACT
This chapter examines the device drivers for training accelerators, prediction accelerators, and DMA devices in the Linux operating system environment. It also examines the device driver for the training accelerator control unit and the predictor accelerator execution unit controller by using the Linux character device driver framework model. For the DMA device driver, the chapter explores the character device driver framework model and memory mapping technique. It analyzes the training phase of the three algorithms: user-based collaborative filtering (CF), item-based CF, and SlopeOne, which may mine the computational hotspots and common features. The chapter investigates and summarizes the research on hardware acceleration of machine learning algorithms. It also summarizes the hardware acceleration methods that are frequently used in the hardware acceleration research work. These methods are mainly divided into two aspects: speeding up the computation process and reducing the communication cost.
