Novel quantum inspired binary neural network algorithm OM PRAKASH PATEL * and ARUNA TIWARI Department of Computer Science and Engineering, Indian Institute of Technology Indore, Indore 453552, India e-mail: oppatel13@gmail.com; artiwari@iiti.ac.in MS received 16 April 2015; revised 5 April 2016; accepted 5 June 2016 Abstract. In this paper, a quantum based binary neural network algorithm is proposed, named as novel quantum binary neural network algorithm (NQ-BNN). It forms a neural network structure by deciding weights and separability parameter in quantum based manner. Quantum computing concept represents solution proba- bilistically and gives large search space to find optimal value of required parameters using Gaussian random number generator. The neural network structure forms constructively having three number of layers input layer: hidden layer and output layer. A constructive way of deciding the network eliminates the unnecessary training of neural network. A new parameter that is a quantum separability parameter (QSP) is introduced here, which finds an optimal separability plane to classify input samples. During learning, it searches for an optimal separability plane. This parameter is taken as the threshold of neuron for learning of neural network. This algorithm is tested with three benchmark datasets and produces improved results than existing quantum inspired and other clas- sification approaches. Keywords. Quantum computing; neural network; quantum gates; classification; separability plane. 1. Introduction Artificial neural networks have been successfully applied to problems in pattern classification, pattern matching, asso- ciative memories, optimization and function approximation [1–3]. Several architectures have been proposed like per- ceptron, backpropagation, recurrent network, etc. to solve the problem from various fields like mathematics, medi- cine, economics, computer science and many more [4–6]. The performance of neural network in the mentioned area depends upon several parameters such as network archi- tecture, input data, number of neurons, the number of hidden layers, activation function and weights [7–10]. In this paper, with the help of quantum computing concept and constructive formation of neural network, all these param- eters have been optimized. The proposed algorithm uses the quantum computing concept for the selection of weights required to establish the connection at hidden and output layers. Quantum computing concept was, firstly, introduced in classical computing by Narayanan and Moore [11]. The significant work has been done by Han and Kim to solve the knapsack problem using the quantum computing concept with and without termination criteria [12, 13]. Here, qubit q is defined as a smallest unit of information which have better characteristic of the population diversity than other representations. Since qubits are linear superposition of states of probabilistic thus, with the help of Gaussian random generation it gives diversity to select the optimal value of parameters from large subspace. Lu et al [2] proposed an algorithm for optimizing artificial neural net- works by deciding connection weight and architecture through the quantum computing concept. The quantum computing concept has also been used in several applica- tions. Gandhi et al [14] proposed an algorithm to filter EEG signal for brain computer interface. A novel scheme has been proposed by Li and Xu [15] for speech enhancement based on quantum feed-forward neural network, which produces better results than traditional spectral subtraction and Wiener filtering method. Caraiman and Manta [16] proposed an image processing technique using the quantum concept. As network architecture plays an important role in the performance of the system, therefore, selection of an appropriate network architecture is required. There are many algorithms exist which constructively form the net- work architecture. As [17] proposed an algorithm to design neural network architecture constructively for data classi- fication. Huang and Huang [18] proposed a method to bound number of hidden layer neuron in multilayer per- ceptron. Similarly, [19] proposed an algorithm to bound on the number of hidden layer neurons for the binary neural network classifier. It offers a high degree of parallelism in the hidden layer formation. To overcome the issue of finding proper weights and selection of network architecture, recently a quantum inspired binary neural network algorithm is presented by Patel and Tiwari [20, 21]. In this algorithm, the weights of *For correspondence 1299 Sa¯dhana¯ Vol. 41, No. 11, November 2016, pp. 1299–1309 Ó Indian Academy of Sciences DOI 10.1007/s12046-016-0561-0