eduzhai > Applied Sciences > Engineering >

Digit Image Recognition Using an Ensemble of One-Versus-All Deep Network Classifiers

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: In multiclass deep network classifiers, the burden of classifying samples ofdifferent classes is put on a single classifier. As the result the optimumclassification accuracy is not obtained. Also training times are large due torunning the CNN training on single CPU GPU. However it is known that usingensembles of classifiers increases the performance. Also, the training timescan be reduced by running each member of the ensemble on a separate processor.Ensemble learning has been used in the past for traditional methods to avarying extent and is a hot topic. With the advent of deep learning, ensemblelearning has been applied to the former as well. However, an area which isunexplored and has potential is One-Versus-All (OVA) deep ensemble learning. Inthis paper we explore it and show that by using OVA ensembles of deep networks,improvements in performance of deep networks can be obtained. As shown in thispaper, the classification capability of deep networks can be further increasedby using an ensemble of binary classification (OVA) deep networks. We implementa novel technique for the case of digit image recognition and test and evaluateit on the same. In the proposed approach, a single OVA deep network classifieris dedicated to each category. Subsequently, OVA deep network ensembles havebeen investigated. Every network in an ensemble has been trained by an OVAtraining technique using the Stochastic Gradient Descent with MomentumAlgorithm (SGDMA). For classification of a test sample, the sample is presentedto each network in the ensemble. After prediction score voting, the networkwith the largest score is assumed to have classified the sample. Theexperimentation has been done on the MNIST digit dataset, the USPS+ digitdataset, and MATLAB digit image dataset. Our proposed technique outperforms thebaseline on digit image recognition for all datasets.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...