https://sit.iitd.ac.in/show-news/55
Computer Vision, Machine Learning, Vision Foundational Models, Agentic AI systems for healthcare and mobility, Visual Grounded Reasoning, Visual Recognition, Fairness of AI Models (Chetan Arora)
We are offering several projects for PhD and MSR thesis which will give you an opportunity to drive the future of Healthcare and Mobility. This is an opportunity to be part of a team pushing the frontiers of AI with cutting-edge work in computer vision, medical imaging, and autonomous systems. Our innovations power breakthrough applications in cancer detection, surgical video intelligence, autonomous driving, and advanced driver-assistance systems (ADAS). We publish consistently in the world’s leading venues, including CVPR and MICCAI. Students in our group gain the opportunity to not only publish at top-tier conferences, but also create tangible real-world impact through large-scale deployments with premier partners such as AIIMS Delhi and PGI Chandigarh in healthcare, and Honda, Yamaha, and others in mobility.
Necessary background: Conceptual and hands-on proficiency in modern computer vision techniques and architectures, such as Unsupervised pretraining, Semi-supervised learning, Reinforcement learning, GRPO, DPO, Transformers, Diffusion, DETRs, CLIP, DINO, GDINO, RAG, and agentic frameworks.
Predoc: yes
More details: https://aih-iitd.github.io/
Memory-Efficient Real-Time Genomic Classification from Nanopore Reads on Heterogeneous Edge Hardware(Kolin)
Nanopore sequencing generates continuous, variable-length reads at rates that overwhelm conventional bioinformatics pipelines — basecalling on GPU, taxonomic classification via k-mer databases, and downstream assembly are designed as offline batch processes that assume abundant memory and compute, making them unsuitable for point-of-care diagnostics, field pathogen surveillance, or resource-constrained clinical settings. This problem is to develop a co-designed streaming pipeline targeting a real-time classification latency under 60 seconds from raw signal to species-level call at under 30W system power
Necessary Background: Interest in Sequencing, Computer Architecture and Systems
Using Mobile-Application of Large Language Models for Diagnosis and possible Therapy (Rahul Garg)
We are working towards utilizing large language models (LLMs) for diagnosis and possible treatment of Post Stroke Non-Fluent Aphasia (PSNFA). As a part of earlier work, we have developed a Hindi-adaptation of Melodic Intonation Therapy for treatment of Post Stroke Non-Fluent Aphasia (PSNFA), with very promising outcomes. This work was done in collaboration with AIIMS Delhi and other hospitals (see our paper in Aphasiology, the top journal in this field). We are now evaluating if LLM-based mobile applications can be useful for diagnosis and possible treatment of this condition and other medical conditions.
Desired Qualifications: Good grasp on Machine Learning. Excellent programming and problem solving skills. Development of Mobile applications. Experience of working with LLMs. Interest in Biology and medical applications.
MSR position preferred. A PhD position is also ok.
Predoc: yes
Machine Learning Applications of Physiological Data Analysis (Rahul Garg)
This project is being done in collaboration with AIIMS Delhi. The project aims to develop open source reproducible and medical grade machine learning / deep learning models for common applications in the medical domain. The datasets include Electrocardiography (ECG), Photoplethysmography (PPG) amongst others. The types of problems include medical grade detection of Atrial Fibrillation using ECG, and/or PPG, cardiac rhythm monitoring problems, non-invasive blood glucose estimation in healthy and diseased population, stroke-type detection using physiology data, open source physiology data collection platform.
Desired Qualifications: Good grasp on Machine Learning and Deep Learning. Excellent programming and problem solving skills. Development of Mobile applications. Interest in Biology and medical applications.
MSR position preferred. A PhD position is also ok.
Predoc: yes