Soumava Paul

Hi, visitor! This is Soumava and you have reached my small corner on the World Wide Web. I am an aspiring computer scientist from the city of Kolkata, India. Currently, I am finishing up with my undergrad at Indian Institute of Technology Kharagpur. I will be graduating with a major in Electrical Engineering and a minor in Computer Science and Engineering.

Previously, I was a research intern at IBM India Research Labs, Bangalore where I worked on novel zero-shot learning algorithms. For my Bachelor's Thesis, I have worked on projects from two different areas - Surgical Video Analytics under the supervision of Prof. Debdoot Sheet (Autumn'19), and Singing Voice Detection under the supervision of Prof. K. Sreenivasa Rao (Spring'20).

Email  /  Github  /  Google Scholar  /  ResearchGate  /  LinkedIn  /  Twitter

profile photo

Broadly, I am interested in computer vision, machine learning, and music information retrieval. Below are some of my major projects during my 4 years as an undergrad.

Exploring Knowledge Distillation Techniques in Singing Voice Detection
Soumava Paul, Gurunath Reddy M, K. Sreenivasa Rao
Bachelor's Thesis

Knowledge Distillation with state-of-the-art voice detection models as teachers can substantially boost performance of student models upto 1000x smaller in parameter count.

Addressing Target Shift in Zero-shot Learning using Grouped Adversarial Learning
Saneem Ahmed Chemmengath*, Soumava Paul*, Samarth Bharadwaj, Suranjana Samanta, Karthik Sankaranarayanan
arXiv preprint

Grouped Adversarial Learning (gAL) can reduce effects of target shift in zero-shot learning algorithms.

Jointly Learning Convolutional Representations to Compress Radiological Images and Classify Thoracic Diseases in the Compressed Domain
Ekagra Ranjan*, Soumava Paul*, Siddharth Kapoor, Aupendu Kar, Ramanathan Sethuraman , Debdoot Sheet
ICVGIP, ACM, 2018   (Oral Presentation)

Downscaling high resolution Chest X-Ray Images using an autoencoder (instead of usual interpolation techniques) leads to superior retention of important pathological features for thoracic disease classification.

This, is more classy