Our Projects

Goal and Objectives or brief

The research group aims to leverage cutting-edge computational methods to address critical challenges in health care. We focus on developing advanced techniques for medical image segmentation, explore innovative methods for medical image reconstruction, address disease classification issues by employing machine learning models for accurate diagnosis and prognosis. We also work on developing predictive models using AI techniques to identify risk factors for chronic diseases by integrating and analyzing multi-modal data.

Deep  Neural  Networks(DNN) have become popular for various applications in the domain of image and computer vision due to their well-established performance attributes. DNN algorithms involve powerful multilevel feature extractions resulting in an extensive range of parameters and memory footprints. However, memory bandwidth requirements, memory footprint and the associated power consumption of models are issues to be addressed to deploy DNN models on embedded platforms for real time vision-based applications. We use transfer learning approach to propose resource optimized architecture for porting on to edge devices.

Video classification is used in fields such as surveillance, entertainment, and autonomous driving. Video classification models pose challenges for deployment on resource-constrained devices and this study focuses on optimising Deep Neural Networks (DNN) for computation and memory-efficient video classification. DNN models are optimised using an approach called Stripe-Wise-Pruning (SWP) SWP is a parameter removal method that selectively removes stripes within the filters in the convolutional layers of DNNs based on their importance. In order to maximise computational and memory efficiency, the SparPen optimizer is used with SWP. Key findings of this study indicate that SWP significantly reduces the computational cost and memory utilisation of DNN-based video classification models without compromising their accuracy.