I am a PhD researcher at IIT Delhi, working on World Modelling and SE(3) Equivariant Learning for molecular generation and simulation. I am honored to be a Prime Minister's Research Fellow (PMRF), India's most prestigious PhD fellowship.
My research focuses on developing scalable approaches to learning equivariance in molecular systems. Traditional architectures that hardcode equivariance struggle to scale to larger molecules like proteins or handle complex molecular tasks. My work explores learning equivariance from randomly rotated molecules using world modelling techniques, training models with over 170 million parameters.
Challenge: Hardcoding equivariance into architectures limits scalability to larger molecules (like proteins) and complex molecular tasks. Learning equivariance from randomly rotated molecules is essential.
Approach: I develop a solution using point cloud representations of molecules voxelized with Gaussian noise. The system employs:
Scale: Successfully trained a 170 million parameter model for this task, demonstrating the approach's scalability and effectiveness.
Developed a two-stage architecture: (1) Encoder-decoder with slot attention producing 11 slots trained on CLEVERTex dataset, (2) Diffusion model with U-Net architecture using cross-attention with slots for conditional image generation.
Implemented sequence-to-sequence translation with attention mechanisms, comparing GloVe and BERT embeddings. LSTM decoder used for generating mathematical expressions from natural language text.
Built ResNet architecture (VGG16-inspired) for bird species classification. Explored various normalization techniques (Batch, Group, Layer, Instance, Batch-Instance) and generated attention maps using Grad-CAM for explainability.
Indian Institute of Technology, Delhi
CGPA: 8.10/10.0
Research in World Modelling and SE(3) Equivariant Learning for molecular systems.
Uzio
Indian Institute of Technology (ISM), Dhanbad
CGPA: 8.40/10.0
Awarded India's most prestigious PhD fellowship worth over $70,000 USD for 5 years at IIT Delhi. Recognizes exceptional research potential and academic excellence.
Received the Copper Club Scholarship worth $10,000 USD to fund undergraduate studies at IIT Dhanbad.
Secured 99.8+ percentile among 1.3 million students in India's most competitive engineering entrance examination to gain admission to IIT Dhanbad.
Presented world modelling research at Nepal AI School, sharing insights on SE(3) equivariant learning with the international AI research community.
I'm always interested in discussing research collaborations, innovative projects, or opportunities in deep learning and molecular modelling.
