Browsing by Author "Chinglemba, Thounaojam"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Neural Architecture Search Using Nature Inspired Algorithm For Image Classification
    (Department of Computer Science & Engineering, NITS, 2022) Chinglemba, Thounaojam
    Over the years CNNs models are used effectively and extensively with remarkable accuracy for the image classification task. As newer CNN architecture can classify images with great accuracy, the architecture has become complex and difficult to tune and implement. Over the past four to five years, renewed and breakthrough progress has been made to develop models called Neural Architecture Search (NAS) models which can automatically develop new CNN architecture with fits our classification problems. By developing these NAS models, people with less knowledge about CNN architecture will be able to generate CNN architecture that fits their classification problem. As we try to develop new Neural Architecture Search models using traditional techniques like brute force, it takes huge computational time and power to develop and run these models. So, to develop better NAS models, we look to nature-inspired algorithms as these algorithms tend to develop collective intelligence and self-organize themselves even in the most complex environment to obtain a common solution. For this thesis work, two NAS models are proposed based on the PSO algorithm and compare it other states of the art NAS models and manually tuned CNN architectures.