A Shallow Convolutional Neural Network for Accurate Handwritten Digits Classification
At present the deep neural network is the hottest topic in the domain of machine learning and can accomplish a deep hierarchical representation of the input data. Due to deep architecture the large convolutional neural networks can reach very small test e
- PDF / 1,035,820 Bytes
- 9 Pages / 439.37 x 666.14 pts Page_size
- 73 Downloads / 249 Views
1
Brest State Technical University, Moskowskaja 267, 224017 Brest, Belarus [email protected] 2 Research Institute for Intelligent Computer Systems, Ternopil National Economic University, 3 Peremoga Square, Ternopil, 46020, Ukraine [email protected]
Abstract. At present the deep neural network is the hottest topic in the domain of machine learning and can accomplish a deep hierarchical representation of the input data. Due to deep architecture the large convolutional neural networks can reach very small test error rates below 0.4% using the MNIST database. In this work we have shown, that high accuracy can be achieved using reduced shallow convolutional neural network without adding distortions for digits. The main contribution of this paper is to point out how using simplified convolutional neural network is to obtain test error rate 0.71% on the MNIST handwritten digit bench‐ mark. It permits to reduce computational resources in order to model convolu‐ tional neural network. Keywords: Convolutional neural networks · Handwritten digits · Data classification
1
Introduction
An artificial neural network is powerful tool in different domains [1–5]. Over the last decade the machine learning techniques has the leading role in domain of artificial intelligence [1]. This is confirmed by recent qualitative achievements in images, video, speech recognition, natural language processing, big data processing and visualization, etc. [1–18]. These achievements are primarily associated with new paradigm in machine learning, namely deep neural networks and deep learning [2, 6–18]. However in many real world applications the important problem is limited computational resources, which doesn’t permit to use deep neural networks. Therefore the further development of shallow architecture is an important task. It should be noted especially that for many real applications, the shallow architecture can show the comparable accuracy in compar‐ ison with deep neural networks. This paper deals with a convolutional neural network for handwritten digits classi‐ fication. We propose a simplified architecture of convolutional neural networks, which permits to classify handwritten digits with more precision than a conventional convo‐ lution neural network LeNet -5. We have shown that by using a simplest convolutional neural network can be obtained the better classification results. © Springer International Publishing AG 2017 V.V. Krasnoproshin and S.V. Ablameyko (Eds.): PRIP 2016, CCIS 673, pp. 77–85, 2017. DOI: 10.1007/978-3-319-54220-1_8
78
V. Golovko et al.
The rest of the paper is organized as follows. Section 2 introduces the standard convolutional neural networks. In Sect. 3 we propose a simplified convolutional network. Section 4 demonstrates the results of experiments and finally Sect. 5 gives conclusion.
2
Related Works
Convolutional neural network is a further development of a multilayer perceptron and neocognitron and is widely used for image processing [19–22]. This kind of neural network is invariant to shifts and distortions of the input. Conv
Data Loading...
