In this paper we show that capsule network with some changes in its architecture and with the help of dynamic routing can mimic the speech processing section of the brain to some extent. The results obtained are state of the art and it also challenges some aspects of the capsule network architecture proposed by [Hinton et al., 2017]. This paper also makes a few changes in the selection procedure of the N-gram model proposed by [Wei Zhao et al., 2018]. The paper proposes the idea of mimicking the brain architecture for speech recognition using capsule network by clustering the final capsules into groups of similar lengths of vectors which may represent a specific section of the brain to understand properties of a text. As a result the instantiation properties of text are not lost.
Madhuram, M., Dasgupta, M., Aqib Muhammed Ashik, B. T., & Surya, M. (2019). Clustered capsule network architecture for text classification. International Journal of Engineering and Advanced Technology, 8(5), 225–227.