NLP

Large-Scale Visual Object Discovery through Diffusion Attentive Attribution Text2Image Heatmaps using Stable Diffusion. Python package - Generate OD Synthetic Data in 30 lines of code!

Contextual Information-rich joint embedding for image and text in a multi-modal vector space using object-text collocation and Relative Position-based Transformer

Scanned Document Representation Learning using Image-Text-Loc Fusion CNN-Transformer and leveraging it for clustering and classification into 16 document categories

A one month-long summer school on Advanced Natural Language Processing with special focus on Speech and Machine Translation for low-resourced Indian Languages and MT evaluation techniques.

Explored several baseline Transformer models built from scratch for English to low-resourced North-Eastern Indian Languages in an effort to mainstream large native population