Short Bio
I recently completed my Ph.D. under the supervision of Prof. Srikanta Bedathur and Prof. Prem Kumar Kalra at the Amar Nath and Shashi Khosla School of Information Technology at IIT Delhi.
Research Interests:
1. Multimodal Learning - Vision Language Modes
1. Generative AI - Code LLMs, VLMs
1. Topological Deep Learning
2. Copmuter Vision
3. Point Clouds
4. Graphs
Prior to joining IIT Delhi, I worked as a Research Associate in the Machine learning Group at Indian Institute of Science, Bangalore advised by Prof. Chiranjib Bhattacharyya. Even before, I was a Master’s student in Computer Science and Engineering at the Indian Institute of Technology Guwahati. Please visit my Google Scholar profile for more details.
Current Research and Collaborations
1. Calibration of Image retrieval systems for queries with ambiguous negations (with Prof. Srikanta Bedathur)
+ Ambiguous queries - especially multiple negation are challenging for retrieval systems
+ Calibration of retrieval outcomes to balance multiple alternative outcomes as equally likely
2. Topological Alignment of VLMs and 3D foundation models with the real world
+ Study of topological correctness of VLM-generated images and 3D foundation models generated point clouds, w.r.t the real world
+ Use topologically invariant properties to align these foundation models to a realistic representation of the world
3. Investigating the use of Code-LLMs to solve NP-hard combinatorial problems (with Prof. Sayan Ranu)
+ Neural approaches require large amounts of ground truth data, which is NP-hard to compute, operate as black boxes offering limited interpretability. and also lack cross-domain generalization
+ Design self-evolutionalry approaches for generating programs to solve NP-hard problems including end-to-end interpretability with grund truth supervision.
+ Optimizing the order of LLM calls and balancing use of queries
2. Unsupervised User-persona groups discovery using Reinforcement Learning Based Human Feedback (with Prof. Sayan Ranu)
+ Unsupervised persona identification for e-commerce platforms
+ Develop a RLHF reward model to learn implicit persona information via conformal product/category/super-category/subcategory pair agreement
News
June 2025: Our work on Investigating and Characterizing Real-World Point Clouds w.r.t. Topological Properties - Revisiting Point Cloud Completion: Are We Ready For The Real-World?, in collabaration with the University of Antwerp has been accepted at ICCV - 2025
May 2025: Our work on Real World Generalizable Persona Indentification - [Persona Identification in E-Commerce with Scarce Labels and In-Context Graph Learning](https://dl.acm.org/doi/10.1145/3711896.3737080 in collaboration with Prof. Sayan Ranu, Prof. Abhijnan Chakraborty and FLipkart India Pvt. Ltd. has been accepted at KDD Research Track - 2025
June 2024: Our Paper GLiDR: Topologically Regularized Graph Generative Network for Sparse LiDAR Point Clouds invited and presented at the 1st Workshop on Topological Deep Learning for Computer Vision at CVPR-24
May 2024::Our paper on Label-less Segmentation for LiDAR point clouds in constrained settings MOVES: Movable and moving LiDAR scene segmentation in label-free settings using static reconstruction - Differentiable SLAM Helps Deep Learning-based LiDAR Perception Tasks accepted at Pattern Recognition, Volume 155, November 2024
Feb 2024: Our Paper on Topological Deep Learning on Sparse Point clouds - GLiDR: Topologically Regularized Graph Generative Network for Sparse LiDAR Point Clouds accepted at CVPR-2024
Nov 2023::Our paper on Differentiable SLAM for end-to-end-learning pipelines - Differentiable SLAM Helps Deep Learning-based LiDAR Perception Tasks accepted at British Machine Vision Conference (BMVC) - 2023
Nov 2020::Our paper on Adversarial learning for Domain Translation of Point clouds - Dynamic to Static Lidar Scan Reconstruction Using Adversarially Trained Auto Encoder Accepted at AAAI - 2021