Language Classification Technique Using In-memory High Dimensional Computing (HDC)

Abstract

In-memory High Dimensional Computing (HDC) involves generation, storage and operations on High Dimensional (HD) vectors. The idea of High Dimensional Computing is evolved from the working of human brain which process the data from millions of neurons at a time with less power hungry and high classification accuracy. In this work, we propose 2-D architecture for language identification as an application for High Dimensional Computing. 10000 bit high dimensional random binary vectors are used for representing the operands of computing and a ring oscillator based True Random number Generator (TRNG) is used to generate the entity. The experimental results, design details of the architectures, advantages and drawbacks of HDC are also presented in this work. The HD vector computations have high classification accuracy, robust in nature, and have one shot learning as compared to conventional deep learning methods. The current model was validated by identification of 21 European languages. The similar works can also be used for pattern recognition, bio signal processing, voice classification and many more applications. The experiments carried on High Dimensional Computing opens a new paradigm for machine learning applications.