Research Projects

SpeakUp – ML based Speech Aid (Brain Computer Interface)


This study developed a Silent-Speech-Interface (SSI) to assist patients with speech disorders. The SSI enables voiceless communication by capturing electromyography (EMG) signals from the speech system and classifying them in real-time using a trained Machine Learning (ML) model. Among the tested algorithms, the SVM-ML algorithm achieved the highest SSI accuracy at 90.1%.

Overall, this study involves the creation of a device that measures biomedical signals and translates them into speech accurately using an ML algorithm.  The study's findings contribute to the advancement of accurate SSIs by identifying effective ML algorithms for implementation.

Autofocus Eyeglasses using Liquid Lenses

Abstract Summary:

This project aims to develop a device that addresses the vision problems faced by a significant portion of the world's population. By integrating a voltage-controlled liquid lens, the device mimics the functionality of phoropters and multifocal visual aids without the associated drawbacks. Comparative data analysis indicates that the device significantly improves vision compared to commonly used visual aids. The device enables finer adjustments for focusing light on the optic nerve and blocks stray light, resulting in increased visual acuity. Overall, this auto-focal visual aid outperforms traditional visual aids by adapting to the user's specific visual needs, providing a notable improvement in vision.

Eliminating Human-Induced Machine Learning Bias

Abstract Summary:

This research paper explores the topic of algorithmic bias in Artificial Intelligence (AI) systems and proposes potential solutions to mitigate discrimination. The study identifies two categories of solutions: Pre-Processing and In-Processing. This research argues that eliminating the usage of discriminatory variables holds promise in reducing AI discrimination. Further research is needed to effectively address algorithmic bias and promote a more equitable and objective AI landscape.

SolAR Tracking System for Efficient Energy Generation

Abstract Summary:

This project aims to develop a cost-effective and adaptable solar energy harvester. A comparison is made between Tower Based Solar Systems and Light Tracking Solar Systems to determine their efficiency. The hypothesis states that Solar Tracking Systems generate more photovoltaic energy compared to Tower based systems. The Solar Tower consists of a PVC pipe mounted between a 3-D printed prism and base, while the Solar Tracker utilizes two servos in a 3-D printed coaxial holder. Statistical analysis confirmed the significantly higher energy harvested by both the Solar Tracker and the Solar Tower compared to Stationary solar panels.