Paper Info
Title
Investigation on adaptive data condensation for exemplar based method in speech task.
Abstract
Exemplar based methods are attractive for various speech task because of their ability to learn a local model from the training data. These methods can capture fine details in data with small number of parameters to learn the model. In the era of big data, exemplar based methods face key challenge in terms of computational time. One popular method in exemplar based methods used in various speech task is k-nearest neighhbor (k-NN). K-NN computation time is high as it has to compute distance for each test instance with all the intances in the traning data (also known as instance space). The computational time increases with an increase in the size of the training dataset. In this paper, we propopse an adaptive data condensation scheme for k-NN by doing instance space reduction based on the speaker similarity. With our proposed adaptive condensed nearest neighbor approach, we obtained a significant reduction in the size of the instance space and an improvement in computational time by 83.1% at the expense of 1.1% absolute decrease in classification performance.
Year
Venue
Keywords
2017
IEEE Global Conference on Signal and Information Processing
Condensed nearest neighbor,phoneme classification,deep neural networks,instance space reduction,speedup
Field
DocType
ISSN
Small number,Training set,k-nearest neighbors algorithm,Pattern recognition,Condensation,Computer science,Artificial intelligence,Big data,Deep neural networks,Speedup,Computation
Conference
2376-4066
Citations 
PageRank 
References 
0
0.34
0
Authors
2
Name
Order
Citations
PageRank
Muhammad Rizwan1339.89
David V. Anderson241875.23