Analysis of radial basis function circuits for support vector machine classification

Yim, Chris

Permalink

https://hdl.handle.net/2142/97901 Copy

Description

Title

Analysis of radial basis function circuits for support vector machine classification

Author(s)

Yim, Chris

Contributor(s)

• Shanbhag, Naresh
• Gonugondla, Sujan Kumar

Issue Date

2017-05

Keyword(s)

• radial basis function circuits
• support vector machines
• classification

Abstract

Support vector machines (SVMs) are a very popular machine-learning algorithm used in many systems today. In some applications, having the classifier built into a chip can allow for low-power and efficient operation. With this in mind, in this senior thesis multiple radial basis function (RBF) circuits for classification are implemented in a 180-nm-process technology. After evaluating the power, energy, delay, and accuracy of different circuit architectures, the Gilbert Gaussian and a newly proposed complementary bump circuit were shown to be the best for implementing in a support vector machine classifier. The two-dimensional Gilbert Gaussian circuit has the most accurate performances, whereas the newly proposed two-dimensional complementary bump circuit has the smallest area. Moreover, the proposed bump circuit also has smaller energy and power consumption than the Gilbert Gaussian circuit at the same input current levels.

Type of Resource

text

Language

en

Permalink

http://hdl.handle.net/2142/97901

Owning Collections