work Boosting image retrieval with accurately labelled small data The project investigates how multi-labeling images with detailed information would boost representation in an image retrieval task, thereby lowering the compute cost significantly. Distilling Word Embeddings via Contrastive Learning A method to distill or compress word embeddings into a four-layer CNN. Making Vision Transformer faster through layer skipping A layer-skipping dynamic vision transformer (ViT) network that skips layers for each sample based on decisions given by a reinforcement learning agent Scientific Q&A Chatbot A chatbot built using Retrieval Augmented Generation (RAG) that instantly generates answers to questions by scanning large collections of scientific papers. fun