Sumit Agarwal

Graduate Research Assistant. Language Technologies Institute. CMU.

prof_pic.jpg

I am a second year MLT student (Masters in Language Technologies) in the Language Technologies Institute, School of Computer Science at Carnegie Mellon University advised by Prof. Teruko Mitamura.

I have developed a keen interest in the recent advancements in NLP, such as GPT-4, ChatGPT, DALLE, and their ability to comprehend human language and perform equally well across diverse tasks. My research focuses on applying deep learning techniques to a variety of NLP tasks, including question answering, information extraction, code generation, summarization, often in a zero or few-shot setting and across multiple languages and modalities.

Before coming to CMU, I worked in Samsung Research Institute, Bangalore, India for four years where I was working in the Voice Intelligence Team, Bixby, on task oriented dialogue systems. I was a part of the team which was responsible for shipping the voice assistant across multiple domains and devices in a seamless manner. I graduated with a B.Tech. in Computer Science and Engineering from Indian Institute of Technology (IIT), Kharagpur in 2017. In my Bachelor’s thesis under Prof. Niloy Ganguly, I worked on modeling opinion dynamics in social networks using recurrent neural networks.

In my leisure time, I love cooking different recipes from Indian cuisine, especially biryani. I also like to click photographs and watch movies and documenteries.

news

Mar, 23 Our work CodeBERTScore: Evaluating Code Generation with Pretrained Models of Code was accepted as a Spotlight paper at Deep Learning for Code (DL4C) workshop at ICLR’23
Dec, 22 Attended EMNLP’22 in Abu Dhabi, UAE.
Oct, 22 Our work PRO-CS : An Instance-Based Prompt Composition Technique for Code-Switched Tasks was accepted as a long paper at EMNLP’22
Jun, 22 Our work Zero shot cross lingual open domain question answering was accepted at Multilingual Information Access (MIA) Workshop at NAACL’22. Our system stood 3rd on the MIA Shared Task.
May, 22 Attended ACL’22 in Dublin, Ireland.
Apr, 22 Our work R3 : Refined Retriever-Reader pipeline for Multidoc2dial was accepted at DialDoc Workshop at ACL’22. Our system stood 1st on the unseen Multidoc2dial Shared Task.