Shravan Shanmugam

I am a Research Assistant at the Indian Institute of Science, Bangalore, advised by Prof. Pradipta Biswas. I currently work on problems related to Computer Vision, focusing on Diffusion Models.

In parallel, I work on Emotion Analysis in collaboration with Dr. Andrea Kohlhase & Dr. Michael Kohlhase from the University of Applied Sciences Neu-Ulm, Germany.

Before this, I led a research project while completing my Bachelor’s at NMIT, Bangalore advised by Dr. Krishanarao Venkatesh, Dr. Vasanthakumar G U & Dr. Dileep Reddy Bolla in colloboration with Unisys.

Research

Beyond my work in human emotion analysis, I work at the intersection of natural language processing and generative AI. I leverage LLMs powered by Natural Language to enhance diffusion models, exploring novel ways to generate creative and visually compelling images and videos.

Diffuse Your Data Blues: Augmenting Low-Resource Datasets via User-Assisted Diffusion

Yashaswi Sinha, Shravan Shanmugam*, Yash Kumar Sahu*, Abhishek Mukhopadhyay, Pradipta Biswas

ACM Conference on Intelligent User Interfaces (IUI), 2025 (Accepted)

This work enhances low-resource datasets by allowing users to select various backgrounds and foregrounds, which are seamlessly blended using diffusion models. By ensuring natural integration between elements, the approach improves object detection metrics, offering a novel solution for augmenting datasets in resource-constrained environments.

Intelligent Weather Prediction using Localized Prophet Model

Shravan Shanmugam, Shrasti Katiyar, Simran Runwal, Vachaspati Goutam, Durai Narayanan, S Vasudev, G U Vasanthakumar, Dileep Reddy Bolla, Krishnarao Venkatesh, H Sarojadevi

Intelligent Systems with Applications, 2024 (Under Review)

An embedded system was developed to predict hyper-local weather conditions using TinyML and EdgeAI, enhancing automated farming practices. Edge computing enables real-time decision-making for tasks like irrigation and pesticide spraying without human intervention. Time series models were implemented to improve prediction accuracy, along with a smart alert system that automatically reduces temperature during high-temperature events. Meta's Prophet model was utilized for time series forecasting on a resource-constrained device.

Comparative Analysis of Diverse Multi-object Techniques Utilized for Test Case Prioritization

Dr. Nirmala JS, Shravan Shanmugam, Simran Runwal, Shrasti Katiyar, Syed Abdur Raheman, Shubh Chokani

ICRET, 2024

The comparative analysis focuses on diverse multi-objective techniques employed for test case prioritization, evaluating their effectiveness in optimizing testing processes. By examining various methods, the analysis highlights differences in approach, efficiency, and performance across techniques aimed at improving fault detection, coverage, and execution time. The study provides valuable insights into how these techniques balance trade-offs between multiple objectives, offering guidance on selecting the most suitable approach for different testing scenarios.