Browsing by Author "Fitsum, Assamnew (PhD)"

Now showing 1 - 2 of 2
  • No Thumbnail Available
    Item
    Acceleration of Convolutional Neural Network Training using Field Programmable Gate Arrays
    (Addis Ababa University, 2022-01) Guta, Tesema; Fitsum, Assamnew (PhD)
    Convolutional neural networks (CNN) training often necessitates a considerable amount of computational resources. In recent years, several studies have proposed CNN inference and training accelerators, which the FPGAs have previously demonstrated good performance and energy efficiency. To speed processing, the CNN requires additional computational resources such as memory bandwidth, a FPGA plantform resource usage, time, and power consumption. As well as training the CNN needs large datasets and computational power, and they are constrained by the requirement for improved hardware acceleration to support scalability beyond existing data and model sizes. In this study, we propose a procedure for energy efficient CNN training in collaboration with an FPGA-based accelerator. We employed optimizations such as quantization, which is a common model compression technique, to speed up the CNN training process. Additionally, a gradient accumulation buffer is used to ensure maximum operating efficiency while maintaining gradient descent of the learning algorithm. Subsequently, to validate our design, we implemented the AlexNet and VGG16 models on an FPGA board and a laptop CPU and GPU. Consequently, our designs achieve 203.75 GOPS on Terasic DE1-SoC with the AlexNet model and 196.50 GOPS with the VGG16 model on Terasic DE-SoC. This, as far as we know, outperforms existing FPGA-based accelerators. Compared to the CPU and GPU, our design is 22.613X and 3.709X more energy efficient respectively.
  • No Thumbnail Available
    Item
    Image Deblurring with Compressive Sensing
    (Addis Ababa University, 2021-12) Rahel, Berhanu; Fitsum, Assamnew (PhD)
    Compressive sensing is a technique which enables recovery of signals that are represented by an underdetermined system of equations. Such a recovery of an original signal is made possible if the samples are represented in a sparse manner provided an appropriate measuring matrix is used for the modeled system. Blurred images are examples of signals that are sparse especially in transform domains. Different researches have been done to show the possibility of recovering blurred images that use sparse representation of transform domains by applying compressive sensing. In this thesis, however, a model has been used that doesn’t require transforming into other domains. In addition, a box-wise approach has been used that derives the underdetermined system matrix from 7x7 segmented boxes of the blurred image. Then compressive sensing algorithms are used to recover the whole image iteratively. This method is shown to have a much better computational complexity, for example, than the traditional Lucy-Richardson deblurring but it has limitations due to approximations used in the 7x7 boxes during modeling. Thus, with this improved computational complexity, the study provides an initial platform to deblur images using box-wise method and compressive sensing theories.