Research Objective

I aim to understand how humans solve a problem and I aim to model that behavior and its characteristics into the models that I build. The primary way I attempt to do this in my work is by leveraging multimodal computing. I also aim to try and bring the process of human reasoning in my models by combining NLP with commonsense reasoning and social and psychological cues. For example, in the case of speech emotion recognition, humans have a very defined process of segmenting emotions hierarchically. They do not perceive emotion as a single level decision making process where the original emotion is perceived out of the possible outcomes in one go. Instead, the human process is to segment emotion hierarchically similar to a decision tree. My research objective is to try and model these human cues in my models - Like we tried to model in our first paper by using multimodal inputs to a hierarchical decision tree like structure.

Current Research

I am currently on track to join CLSP (Center for Language and Speech Processing) at Johns Hopkins University. Ranked in the top 5 in this field, I will be working under Prof. Jesus Villalba, this fall, in the field of multimodal speech emotion recogntion while also combining the same with speaker identification and aiming to increase the SOTA in this field.

Research Experience (Labs & Internship)

1. APCL (NSIT, Delhi University - 3 Years) - Worked as an undergraduate researcher under Prof. Rana and Prof. Kumar. I mainly did research on language and speech processing in a multimodal fashion. During the first year, I worked mainly on speech processing and how to recognise emotion from speech. During my second and third year, I worked on statistical techniques in NLP like LDA and Clustering. Under the guidance of Professor Rana and Professor Kumar, I published 3 papers in the field out of which 2 received the award by my University for showing Excellence in Research and publishing in journals with an Impact Factor > 8.

2. Draft Singapore (Research Intern - 1 year) - Developed an intelligent learning assistant that uses an ALBERT-based Machine Reader in conjunction with a Deep Retriever to assist students using D.Kraft’s e-learning platform in queries and information retrieval. Also worked on developing a model to automate the creation of quizzes from documents using a modified T5 transformer. The work is currently pending a patent in India.