Improving downstream task performance in code-mixed setting
Supervisor(s): Prof. Dr. Ponnurangam Kumaraguru
Description: We work on improving task performance in code-mixed settings to address challenges of code-mixed data scarcity, by exploring multi-turn language interpretability of PLMs and LLMs, transfer learning techniques, etc.
Temporal reasoning and future event prediction
Supervisor(s): Prof. Dr. Adam Jatowt
Description: We aim to understand the ability of LLMs in temporal reasoning, predicting future events with or without a given context, prompt designing and improving event prediction in general.
Leveraging multimodal data for healthcare and clinical applications
Supervisor(s): Prof. Dr. Sriparna Saha, Abishek Tiwari
Description: We focus on learning from text, image, speech and video for automatic disease diagnosis, response and report generation, clinical conversation summarization and other healthcare applications.
The effect of improving isotropy of embeddings in downstream tasks across domains
Supervisor(s): Dr. Venkatesh Vinayakarao
Description: Our experiments involve finding out whether increasing isotropy of embeddings improve the quality of generation and prediction across tasks in various domains.