Research & Projects

Publications

• ViBe: A Text-to-Video Benchmark for Evaluating Hallucination in Large Multimodal Models (2024) Vipula Rawte, Sarthak Jain*, Aarush Sinha*, Garv Kaushik*, Aman Bansal*, Prathiksha Rumale Vishwanath*, Samyak Rajesh Jain, Aishwarya Naresh Reganti, Vinija Jain, Aman Chadha, Amit P. Sheth, Amitava Das (* Denotes requal contribution)

• Efficient Biomedical Information Retrieval via Bio-Informed Generative Pseudo Labelling(2024) Paper Under Review

Projects

• DocQA: A custom 13.2M parameter cross encoder QA model that gives a diagnosis based on the input of symptoms a user describes in natural language, trained on 24 disease symptoms. An accuracy of 94.5% was achieved in predicting the disease from the symptoms.

• Learning to Forget: This project aims to create a model that learns to forget by randomizing the weights of a custom-trained text classifier on a wine-review dataset. The original model achieved a 90% accuracy, and after randomizing the weights and evaluating the model again, the accuracy dropped to 50%.

• PoincareEmbeddings: My implementation of the paper Poincaré Embeddings for Learning Hierarchical Representations on the dataset HyperLex: A Large Scale Evaluation of Graded Lexical Entailment for learning hierarchical representations in hyperbolic space.