Home > Center for Artificial Intelligence Research (CAIR) > Our Projects

Our Projects

  • CAIR

Engineering Intelligence for The Real World

At CAIR, we’re building a vibrant research community committed to solving real-world problems with intelligent technologies. Our focus spans Edge AI, High-Performance Computing (HPC), Data Science, and Cloud Computing — powering innovation that’s fast, reliable, and future-ready. With strong partnerships across industry and academia, CAIR is where fundamental research transforms into applied breakthroughs.
Engineering Intelligence
  • EDGE AI PROJECTS

Smarter Models. Lighter Systems.

Edge AI enables real-time intelligence closer to the source—on devices where resources are limited but responsiveness is critical. Our research in this space focuses on making powerful AI models lighter, faster, and more adaptable. From medical diagnostics to industrial automation, each project tackles a real-world problem through optimized architectures, advanced pruning techniques, and hardware-efficient deployment.
  • Redundancy & Quantization-Based Optimization of DNNs
  • This project explores advanced deep learning techniques to address key healthcare challenges—ranging from medical image segmentation to disease classification. The team works on compressing and refining DNN architectures by reducing redundancy and applying quantization, enabling faster processing and lower energy consumption on edge devices. The models are designed to analyze multi-modal health data for improved diagnosis, reconstruction, and risk prediction.
  • Transfer Learning for DNN Model Optimization
  • Deep Neural Networks (DNNs) are powerful but often too large for real-time, embedded systems. This project leverages transfer learning to adapt high-performing DNNs into compact, resource-optimized versions that can run efficiently on edge platforms. The result: significant reductions in memory footprint and power usage, making these models viable for real-time vision applications in constrained environments.
  • Filter Pruning for Efficient Video Classification
  • In video classification tasks—from surveillance to autonomous navigation—latency and computational load are major challenges. This research introduces a filter pruning strategy called Stripe-Wise-Pruning (SWP), combined with the SparPen optimizer, to selectively remove less significant filter elements in convolutional layers. The approach delivers highly efficient models with reduced computation and memory usage—without compromising performance.
  • Industry Collaboration: Intelligent Assembly Line at Spicer-DANA
  • In partnership with Spicer-DANA, we developed an AI-driven solution for part validation on assembly lines. The challenge: manual inspection errors impacting productivity. The solution: an optimized computer vision model deployed on Raspberry Pi devices, integrated directly into the manufacturing process. This collaboration not only improved accuracy and consistency but also showcased how lightweight AI can drive industrial innovation at scale.
  • HPC PROJECTS

Compute Power. Scientific Precision.

At CAIR, our High-Performance Computing (HPC) research focuses on tackling large-scale scientific and real-time challenges through deep learning and advanced hardware integration. We apply HPC innovations in key domains such as 3D modeling, hyperspectral imaging, computational fluid dynamics (CFD), and quantum computing.

These initiatives are backed by major government missions such as the National Supercomputing Mission (NSM) and the National Quantum Mission, in partnership with Karnataka Medical College, SDM University, and IIIT Dharwad.

  • Quantum Software Development Lifecycle for Diabetic Retinopathy
  • Currently underway from 2023 to 2025, this MeitY-QCAL-sponsored project focuses on building a structured Quantum Software Development Life Cycle (QSDLC) specifically for diabetic retinopathy classification. With a grant of $19,860, the research is led by Principal Investigator Dr. Satyadhyan Chickerur and Co-Principal Investigator Dr. K M M Rajeshekariah. The project aims to bridge the gap between classical and quantum computing by creating a reliable process for translating classical AI models into quantum-compatible applications, ultimately enhancing diagnostic accuracy and computational efficiency in retinal disease detection.
  • Automated Grading of Diabetic Retinopathy at PHCs
  • Completed between 2021 and 2023, this project was funded by the DST–NSM with a grant of ₹19.886 lakhs. Spearheaded by Dr. Satyadhyan Chickerur, along with co-investigators Dr. Mahesh Patil and Dr. Shantala Giraddi, the study developed a distributed deep learning framework to automate the grading of diabetic retinopathy. Leveraging transfer learning, the research successfully transferred optimal hyperparameters (OHPs) from the widely used EyePACS dataset to the Indian Diabetic Retinopathy Image Dataset (IDRiD), eliminating the need for intensive hyperparameter tuning. This innovation allows scalable deployment across Primary Healthcare Centers, offering faster, more accurate diagnostics in rural and underserved regions.
  • Blood Flow Analysis using Deep Learning and CFD
  • This project merges MRI image processing with computational fluid dynamics to analyze blood flow patterns for cardiovascular diagnostics. A deep learning model extracts accurate vascular flow data, which is then simulated via CFD tools to model hemodynamic conditions with greater precision.
  • Power-Efficient 3D Rendering on Multi-Core Processors
  • Addressing the energy demands of modern rendering pipelines, this study examines GPU power usage during 3D rendering for VR, gaming, and simulations. By analyzing variables like shader calls and scene complexity, the research identifies strategies to optimize GPU performance while lowering energy consumption.
  • Deforestation Monitoring using Hyperspectral Imagery
  • Using a semi-supervised deep learning framework with semantic segmentation, this project identifies and tracks deforestation over time using hyperspectral data. Its fine-grained detection capabilities support environmental change monitoring at a highly detailed level, crucial for conservation and planning efforts.
  • DATA SCIENCE PROJECTS

Meaningful Data. Measurable Impact.

CAIR’s data science research harnesses the power of artificial intelligence to create clinically relevant tools that improve diagnostics, monitoring, and care delivery. From cardiovascular imaging and neonatal health to assistive applications for visual impairments, these projects showcase how data-driven models can transform healthcare. Our work is built in collaboration with KLE Kore Hospital, JNMC KAHER, and Hubli Scan Center.

  • Coronary Artery Plaque Detection from 3D CCTA Images
  • In collaboration with KLE Kore Hospital, Belagavi, this project focuses on automating the segmentation of coronary arteries and detecting plaques from 3D Coronary Computed Tomography Angiography (CCTA) images. The model aims to accurately isolate coronary vessels and assess their structural characteristics—helping clinicians identify early signs of cardiovascular disease, improve treatment outcomes, and reduce diagnostic ambiguity.
  • Kangaroo Mother Care (KMC) Monitoring – KANGA Suraksha
  • Developed in partnership with JNMC KAHER, this project introduces a real-time monitoring system designed to track compliance with Kangaroo Mother Care (KMC) practices. Aimed at improving the survival and health of premature and low birthweight infants, the KANGA Suraksha device reinforces KLE's commitment to neonatal care by integrating technological innovation with maternal-infant wellness protocols.
  • Cortical Visual Impairment (CVI) Support Application
  • This application, created alongside KLE Kore Hospital, is designed to enhance visual engagement and independence for individuals with Cortical Visual Impairment. It functions as a therapy-aligned, user-centric tool that bridges the gap between clinical care and daily functionality—empowering children and adults alike, and supporting educators, therapists, and families in the process.
  • Metal Artifact Reduction (MAR) in CT/MRI Scans
  • In collaboration with Hubli Scan Center, this project addresses the challenge of visual distortion in CT and MRI scans caused by metallic objects such as implants or prosthetics. By refining imaging techniques, the MAR system enhances diagnostic clarity, supporting more accurate clinical interpretations and safer, more effective treatment planning.
  • Low-Dose CT Image Enhancement using FreeSeed Model
  • This project aims to mitigate the risks of radiation exposure in CT imaging by using a dual-domain enhancement model. The FreeSeed network, which combines frequency-based and sinogram-aware self-guided learning, helps recover missing anatomical detail in low-dose, sparse-view CT scans—balancing patient safety with high diagnostic fidelity.
  • PREDICTIVE ANALYTICS

Forecasting with Intelligence

At CAIR, predictive analytics bridges machine learning with actionable foresight. Our models don’t just interpret historical data—they uncover patterns and predict outcomes that drive real-world decisions in agriculture and environmental sciences. These tools enable researchers, farmers, and policymakers to respond proactively and effectively.
  • Late Leaf Spot Detection and Groundnut Pod Quality Assessment
  • In collaboration with the University of Agricultural Sciences, Dharwad, this project develops a deep neural network (DNN) model to detect late leaf spot—a fungal disease affecting groundnut plants. The model not only identifies disease presence but also establishes its impact on pod quality. By correlating leaf disease progression with yield degradation, the system provides critical insights for improving crop health monitoring, enabling early intervention, and guiding informed farming decisions.
  • RESEARCH FOCUS AREAS

Domains of Impact

01

Edge AI

02

High Performance Computing

03

Data Science

04

Cloud Computing