Enhancing the performance of the neural network model for the EMG regression case using Hadamard product

Although neural networks have revolutionized various fields, their deployment in mobile environments confronts significant challenges due to limitations in battery and cooling capacity, especially for internet of things devices [1]. A lightweight neural network is urgently needed to address these issues. In this study, we explore the use of the Hadamard product and assess the usefulness of the method to enhance the neural network performance in mobile environments. The method has less computational complexity compared with other matrix multiplication methods [2]. Hadamard product methodologies are applied to the input features to amplify useful data and diminish noise. Our research involved the use of 48 electromyography signals sourced from the calves of three individuals with the signal time frame being iterated from 10 to 990 with a step size of 10. Findings indicate that the utilization of Hadamard products significantly improves the model performance relative to the increase in model size.