Optimization of deep network models through fine tuning

Purpose–Many strategies have been put forward for training deep network models,however,stacking of several layers of non-linearities typically results in poor propagation of gradients and activations.The purpose of this paper is to explore the use of two steps strategy where initial deep learning model is obtained first by unsupervised learning and then optimizing the initial deep learning model by fine tuning.A number of fine tuning algorithms are explored in this work for optimizing deep learning models.This includes proposing a new algorithm where Backpropagation with adaptive gain algorithm is integrated with Dropout technique and the authors evaluate its performance in the fine tuning of the pretrained deep network.Design/methodology/approach–The parameters of deep neural networks are first learnt using greedy layer-wise unsupervised pretraining.The proposed technique is then used to perform supervised fine tuning of the deep neural network model.Extensive experimental study is performed to evaluate the performance of the proposed fine tuning technique on three benchmark data sets:USPS,Gisette and MNIST.The authors have tested the approach on varying size data sets which include randomly chosen training samples of size 20,50,70 and 100 percent from the original data set.Findings–Through extensive experimental study,it is concluded that the two steps strategy and the proposed fine tuning technique significantly yield promising results in optimization of deep network models.Originality/value–This paper proposes employing several algorithms for fine tuning of deep network model.A new approach that integrates adaptive gain Backpropagation(BP)algorithm with Dropout technique is proposed for fine tuning of deep networks.Evaluation and comparison of various algorithms proposed for fine tuning on three benchmark data sets is presented in the paper.

Structure identification and IO space partitioning in a nonlinear fuzzy system for prediction of patient survival after surgery

Purpose-As far as the treatment of most complex issues in the design is concerned,approaches based on classical artificial intelligence are inferior compared to the ones based on computational intelligence,particularly this involves dealing with vagueness,multi-objectivity and good amount of possible solutions.In practical applications,computational techniques have given best results and the research in this field is continuously growing.The purpose of this paper is to search for a general and effective intelligent tool for prediction of patient survival after surgery.The present study involves the construction of such intelligent computational models using different configurations,including data partitioning techniques that have been experimentally evaluated by applying them over realistic medical data set for the prediction of survival in pancreatic cancer patients.Design/methodology/approach-On the basis of the experiments and research performed over the data belonging to various fields using different intelligent tools,the authors infer that combining or integrating the qualification aspects of fuzzy inference system and quantification aspects of artificial neural network can prove an efficient and better model for prediction.The authors have constructed three soft computing-based adaptive neuro-fuzzy inference system(ANFIS)models with different configurations and data partitioning techniques with an aim to search capable predictive tools that could deal with nonlinear and complex data.After evaluating the models over three shuffles of data(training set,test set and full set),the performances were compared in order to find the best design for prediction of patient survival after surgery.The construction and implementation of models have been performed using MATLAB simulator.Findings-On applying the hybrid intelligent neuro-fuzzy models with different configurations,the authors were able to find its advantage in predicting the survival of patients with pancreatic cancer.Experimental results and comparison between the constructed models conclude that ANFIS with Fuzzy C-means(FCM)partitioning model provides better accuracy in predicting the class with lowest mean square error(MSE)value.Apart from MSE value,other evaluation measure values for FCM partitioning prove to be better than the rest of the models.Therefore,the results demonstrate that the model can be applied to other biomedicine and engineering fields dealing with different complex issues related to imprecision and uncertainty.Originality/value-The originality of paper includes framework showing two-way flow for fuzzy system construction which is further used by the authors in designing the three simulation models with different configurations,including the partitioning methods for prediction of patient survival after surgery.Several experiments were carried out using different shuffles of data to validate the parameters of the model.The performances of the models were compared using various evaluation measures such as MSE.

Novel Adder Circuits Based On Quantum-Dot Cellular Automata (QCA)

Quantum-dot cellular automaton (QCA) is a novel nanotechnology that provides a very different computation platform than traditional CMOS, in which polarization of electrons indicates the digital information. This paper demonstrates designing combinational circuits based on quantum-dot cellular automata (QCA) nanotechnology, which offers a way to implement logic and all interconnections with only one homogeneous layer of cells. In this paper, the authors have proposed a novel design of XOR gate. This model proves designing capabilities of combinational circuits that are compatible with QCA gates within nano-scale. Novel adder circuits such as half adders, full adders, which avoid the fore, mentioned noise paths, crossovers by careful clocking organization, have been proposed. Experiment results show that the performance of proposed designs is more efficient than conventional designs. The modular layouts are verified with the freely available QCA Designer tool.

Mapping cloud computing in university e-governance system

Purpose–The purpose of this paper is to propose a model to map the on-premise computing system of the university with cloud computing for achieving an effective and reliable university e-governance(e-gov)system.Design/methodology/approach–The proposed model incorporates the university’s internal e-gov system with cloud computing in order to achieve better reliability,accessibility and availability of e-gov services while keeping the recurring expenditure low.This model has been implemented(and tested on a university e-gov system)in the University of Kashmir(UOK);case study of this implementation has been chosen as the research methodology to discuss and demonstrate the proposed model.Findings–According to the results based on practical implementation,the proposed model is ideal for e-governed systems as it provided adequate cost savings and high availability(HA)with operational ease,apart from continuing to have the necessary security in place to maintain confidential information such as student details,grades,etc.Practical implications–The implication of this study is to achieve HA and to reduce the cost from using external clouds,mapping internal IT servers of the university with the external cloud computing services.Originality/value–Because no established mapping model for universities has been provided for effective,low-cost,highly available university e-gov system,the proposed mapping model through this paper closes this gap and provides guidelines to implement a hybrid-mapped e-gov model for universities while keeping the recurring expenditure on cloud computing minimal.The paper provides the perceptions of its adoption at UOK for achieving high reliability,accessibility and uptime of its e-gov applications while keeping the recurring expenditure on cloud computing minimal.

Cellular automata-based approach for digital image scrambling

Purpose–The purpose of this paper is to investigate two-dimensional outer totalistic cellular automata(2D-OTCA)rules other than the Game of Life rule for image scrambling.This paper presents a digital image scrambling(DIS)technique based on 2D-OTCA for improving the scrambling degree.The comparison of scrambling performance and computational effort of proposed technique with existing CA-based image scrambling techniques is also presented.Design/methodology/approach–In this paper,a DIS technique based on 2D-OTCA with von Neumann neighborhood(NvN)is proposed.Effect of three important cellular automata(CA)parameters on gray difference degree(GDD)is analyzed:first the OTCA rules,afterwards two different boundary conditions and finally the number of CA generations(k)are tested.The authors selected a random sample of gray-scale images from the Berkeley Segmentation Data set and Benchmark,BSDS300(www2.eecs.berkeley.edu/Research/Projects/CS/vision/bsds/)for the experiments.Initially,the CA is setup with a random initial configuration and the GDD is computed by testing all OTCA rules,one by one,for CA generations ranging from 1 to 10.A subset of these tested rules produces high GDD values and shows positive correlation with the k values.Subsequently,this sample of rules is used with different boundary conditions and applied to the sample image data set to analyze the effect of these boundary conditions on GDD.Finally,in order to compare the scrambling performance of the proposed technique with the existing CA-based image scrambling techniques,the authors use same initial CA configuration,number of CA generations,k紏10,periodic boundary conditions and the same test images.Findings–The experimental results are evaluated and analyzed using GDD parameter and then compared with existing techniques.The technique results in better GDD values with 2D-OTCA rule 171 when compared with existing techniques.The CPU running time of the proposed algorithm is also considerably small as compared to existing techniques.Originality/value–In this paper,the authors focused on using von Neumann neighborhood(NvN)to evolve the CA for image scrambling.The use of NvN reduced the computational effort on one hand,and reduced the CA rule space to 1,024 as compared to about 2.62 lakh rule space available with Moore neighborhood(NM)on the other.The results of this paper are based on original analysis of the proposed work.

A new cluster validity index using maximum cluster spread based compactness measure

Purpose-The most commonly used approaches for cluster validation are based on indices but the majority of the existing cluster validity indices do not work well on data sets of different complexities.The purpose of this paper is to propose a new cluster validity index(ARSD index)that works well on all types of data sets.Design/methodology/approach-The authors introduce a new compactness measure that depicts the typical behaviour of a cluster where more points are located around the centre and lesser points towards the outer edge of the cluster.A novel penalty function is proposed for determining the distinctness measure of clusters.Random linear search-algorithm is employed to evaluate and compare the performance of the five commonly known validity indices and the proposed validity index.The values of the six indices are computed for all nc ranging from(nc_(min),nc_(max))to obtain the optimal number of clusters present in a data set.The data sets used in the experiments include shaped,Gaussian-like and real data sets.Findings-Through extensive experimental study,it is observed that the proposed validity index is found to be more consistent and reliable in indicating the correct number of clusters compared to other validity indices.This is experimentally demonstrated on 11 data sets where the proposed index has achieved better results.Originality/value-The originality of the research paper includes proposing a novel cluster validity index which is used to determine the optimal number of clusters present in data sets of different complexities.