Soumya Chatterjee

Pre-Doctoral Researcher at Google

About Me

I am a Pre-Doctoral Researcher at Google Research.

Previously, I graduated from IIT Bombay with B.Tech in Computer Science and Engineering and a Minor in Statistics. During my undergraduate, I was fortunate to have worked with Prof. Sunita Sarawagi and Preethi Jyothi on the use of word alignments for lexically constrained translation, and with Prof. Ganesh Ramakrishnan and Prof. Soumen Chakrabarti on applications of hierarchical representation learning for text classification and entity typing.

I spent the summer of 2020 at Google Research India working on modelling human behaviour in information sampling tasks. In Summer 2019, I worked with Prof. Wei-Ta Chu on thermal face recognition methods. I have also interned with the RnD team at AWL Inc. in Winter 2019 and the academic content development team at GreyAtom EduTech in Winter 2018.

My research interests include Machine Learning, Natural Language Processing and Representation Learning.

Soumya Chatterjee

Pre-Doctoral Researcher

Google Research


Apr 2022
Paper on improving across-session estimation reliability of behavioural metrics via modelling temporal adjustment of policies to appear at CogSci 2022
Feb 2022
Paper on word alignments and use in lexically contrained translation to appear at ACL 2022
Aug 2021
Graduated from IIT Bombay with a B.Tech in Computer Science
July 2021
Joined Google Research in Bangalore, India as a Pre-Doctoral Researcher
May 2020
Started internship at Google Research India with Pradeep Shenoy
Dec 2019
Started internship at AWL Inc., Sapporo, Japan
Sept 2019
Paper on Thermal Face Recognition to appear at International Conference on Multimedia Modeling, MMM 2020
May 2019
Started internship at National Chung Cheng University with Prof. Wei-Ta Chu
Nov 2018
Started internship at GreyAtom EduTech, Mumbai, India


Accurate Online Posterior Alignments for Principled Lexically-Constrained Decoding
Meta-Learning of Dynamic Policy Adjustments in Inhibitory Control Tasks
Joint Learning of Hyperbolic Label Embeddings for Hierarchical Multi-label Classification
Model-agnostic Fits for Understanding Information Seeking Patterns in Humans
Thermal Face Recognition Based on Transformation by Residual U-Net and Pixel Shuffle Upsampling