Our core research areas, ongoing student projects and funded programs — advancing intelligent computational technologies with real-world impact.
IMPACT Lab conducts interdisciplinary research at the intersection of Artificial Intelligence, Imaging, Data Analytics, and Computational Sciences.
Novel algorithms and intelligent models for real-world decision-making.
Learning-based systems for prediction, classification and compression.
Visual perception for detection, segmentation and scene understanding.
Techniques for enhancing, analyzing and interpreting visual data.
Identifying meaningful structure and regularities within complex data.
Trustworthy, interpretable models that reveal their own reasoning.
AI-driven modelling of photonic and physical sensor systems.
Mathematical and algorithmic models of complex systems and processes.
Extracting actionable insight from large, complex datasets.
Satellite SAR/InSAR analytics for deformation and hazard monitoring.
Intelligent computational tools for diagnosis and biomedical sensing.
Perception and decision-making for self-governing systems.
Lightweight, compressed models built for on-device deployment.
Virtual replicas and intelligent systems for next-gen technologies.
14 completed research projects from IMPACT Lab researchers, built on the areas above.
An interactive 10-step SBAS InSAR workflow built on PyGMTSAR for monitoring millimetre-scale ground deformation from Sentinel-1 SAR data — validated over the Chamoli district, Uttarakhand using 28 scenes and 48 interferograms.
Combines knowledge distillation with Tucker-decomposition-based compression to shrink FOM-prediction models for fiber-optic SPR sensors — retaining high accuracy (R² ≈ 0.98) at a fraction of the parameters for on-device deployment.
A two-stage pipeline for detecting and segmenting landslides from multitemporal Sentinel-2 optical data using physics-informed features and a transformer encoder with curriculum learning — reaching 96.7% detection accuracy.
Proposes Residual-Based Knowledge Distillation with a sliding two-predecessor window to tackle the capacity gap and error-avalanche problems in teacher-assistant distillation chains — outperforming BLKD, TAKD and DGKD on CIFAR-10/100.
Combines LiCSBAS-generated InSAR time-series with GPR, LSTM and GRU deep learning models to forecast volcanic ground deformation in the highly active Campi Flegrei region.
Pairs deep learning with symbolic regression for interpretable modelling of photonic (FO-SPR) sensor performance — improving R² from 0.967 to 0.986 over pure symbolic regression while keeping the model explainable.
Explores new compression techniques — including K-means weight sharing — for MLPs and Kolmogorov-Arnold Networks, achieving up to 35.9× parameter compression on California Housing with virtually no accuracy loss.
An AI-driven Knowledge Distillation framework that transfers knowledge from a complex teacher model to a lightweight student — achieving R² = 98.39% with 4× faster inference for real-time photonic sensing.
Applies physics-informed neural networks to model the 2024–2025 Fentale volcano dyke intrusion, embedding geophysical constraints directly into a deep learning-based deformation model.
Combines spectral analysis with deep segmentation models across multiple remote sensing sources to improve landslide detection generalization across different geographic regions.
Develops an interpretable, physics-aware spiking neural network to optimise oil-water emulsion characterisation using tilted fiber Bragg grating sensor data.
Proposes a hierarchical, uncertainty-aware multimodal learning framework for segmenting debris-covered glaciers from satellite imagery.
Uses InSAR time-series analysis to monitor and characterise land subsidence patterns in the Jharia Coalfield.
Introduces a DA-FNO based deep learning framework for de-noising and accurate glucose prediction from gold-coated tilted fiber Bragg grating (Au-TFBG) optical sensor signals.
Externally funded research and consultancy projects associated with IMPACT Lab.
Research & Development
Research & Development
Lab Development
Consultancy
