Our Projects

Collaborations and Research Projects

Our collaborations and research projects are centered on advancing High-Performance Computing (HPC) capabilities by integrating deep learning techniques with accelerators such as GPUs, QPUs, and FPGAs. Key applications include 3D modeling, hyperspectral imaging, computational fluid dynamics (CFD), and blood flow analysis. These initiatives are supported by sponsored projects under the National Supercomputing Mission (NSM), the National Quantum Mission, and interdisciplinary research programs in collaboration with Karnataka Medical College and the Department of Biomedical Sciences at SDM University. Additionally, we maintain active research partnerships with IIIT Dharwad, fostering innovation across multiple domains.

The Quantum Software Development Life Cycle (QSDLC) framework provides a structured approach to the complex process of quantum application development. Developers often begin with a classical computing mindset, only to realize that quantum computing offers more effective solutions. The QSDLC facilitates this transition by offering a comprehensive lifecycle that enables the integration of classical source code into quantum applications. This lifecycle addresses existing development frameworks' inherent gaps and challenges, guiding developers through seamlessly adapting classical designs to leverage quantum computing capabilities.

Detecting diabetic retinopathy (DR) is a challenging and time-consuming process that typically requires manual analysis of retinal images. Deep learning (DL) models can automate this but require large datasets and significant training time. The performance of these models depends on optimal hyperparameters (OHPs), usually identified through costly hyperparameter optimization (HPO). This study uses transfer learning (TL) to transfer OHPs from the EyePACS DR dataset to the Indian Diabetic Retinopathy Image Dataset (IDRiD). A ResNet model trained on both datasets demonstrates that EyePACS OHPs can be effectively used for IDRiD without additional tuning. This research highlights the reusability of OHPs across different DR datasets, reducing the need for iterative HPO. The optimized hyperparameters for both datasets are provided as starting points for future work.

This research focuses on developing an advanced semi-supervised deep learning framework tailored for detecting deforestation changes over time using semantic segmentation techniques applied to hyperspectral imagery. The framework leverages the unique capabilities of hyperspectral data, which provides detailed spectral information across multiple bands, allowing for more accurate and precise detection of environmental changes, such as deforestation. By employing semantic segmentation, the model can classify and delineate areas of interest within the imagery, identifying regions affected by deforestation at a fine-grained level.