Novel Methods Based on CNN for Improved Bacteria Classification

Recent times have witnessed extensive use of deep learning in both supervised and unsupervised learning problems. One of these models is convolution neural networks (CNN) which has outperformed all others for object recognition and object classification t

  • PDF / 915,214 Bytes
  • 16 Pages / 439.37 x 666.142 pts Page_size
  • 83 Downloads / 182 Views

DOWNLOAD

REPORT


and Rohit Verma2

1 BNY Mellon, 240 Greenwich Street, Manhattan, NY, USA

[email protected] 2 ABV-IIITM, Gwalior, India

Abstract. Recent times have witnessed extensive use of deep learning in both supervised and unsupervised learning problems. One of these models is convolution neural networks (CNN) which has outperformed all others for object recognition and object classification task. Although these convolution neural networks have achieved exceptional accuracies, still a huge amount of iterations create chances of getting stuck in local optima makes it computationally expensive to train. To handle this issue, we have developed some hybrid methods using, particle swarm optimization (PSO), genetic algorithm, and autoencoders for training CNN. We have taken the images of bacteria and classified them into four different genera (E. coli, Listeria, Salmonella, and Staphylococcus) to measure the performances of these models. Through this study, we concluded that evolutionary techniques can be used to train CNN more efficiently. Keywords: Convolution neural networks · Particle swarm · Optimization · Genetic algorithm · Autoencoders

1 Introduction Many of the recent spectacular successes in machine learning involve learning one complex task very well, through extensive training over thousands or millions of training examples [1–3]. After learning is complete, the agent’s knowledge is fixed and unchanging; if the agent is to be applied to a different task, it must be re-trained (fully or partially), again requiring a huge number of new training examples. Deep convolutional networks (CNN) have been crucial to the success of deep learning. Architectures based on CNN have achieved unprecedented accuracy in domains ranging across computer vision [1], speech recognition [4], natural language processing [5–7], and recently even the board game Go [3].

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 X.-S. Yang et al. (eds.), Proceedings of Fifth International Congress on Information and Communication Technology, Advances in Intelligent Systems and Computing 1184, https://doi.org/10.1007/978-981-15-5859-7_1

2

C. Chopra and R. Verma

The performance of deep convolutional networks has improved as these networks have been made ever deeper. For example, some of the best-performing models on ImageNet [8] have employed hundreds or even a thousand layers [9]. However, these extremely deep architectures have been trainable only in conjunction with techniques like residual connections [9] and batch normalization [10]. It is an open question whether these techniques qualitatively improve model performance or whether they are necessary crutches that solely make the networks easier to train. In this work, we developed some hybrid CNN models using autoencoders, PSO, and genetic algorithm individually to optimize the model and improve the classification accuracy. It has already been stated that the CNN networks might be trained end to end in a supervised method while l