About Me

Hello,

I am a fourth year research fellow working towards my PhD under the supervision of Dr. Utpal Garain at the Computer Vision and Pattern Recognition Unit, ISI Kolkata.

My current research is focused on trying to inculcate key features of symbolic AI techniques, like domain knowledge augmentation, logical coherence, etc into deep learning systems. Addition of logical constraints and pre-existing knowledge not only makes these systems more aligned to critical applications but also makes them more data efficient. This problem shows up in many domains and thus leads me to work in several fields like natural language, medicine, biology, astrophysics and diverse business applications.

Best wishes,

Soumadeep Saha


Education

BS & MS

Indian Institute of Science Education and Research, Kolkata

I graduated with a Masters' in Physics and a minor in Mathematics.

MS Thesis

I worked on robust deep learning systems, studying the natural trade-off there is between accuracy and robustness. Given that adversarial training acts like a regularisation term, I looked at various other factors in conjunction with adversarial training to make recommendations about model architectures that strike the best balance between accuracy and robustness.

2015 - 20


Experience

LLM Training Session

At DataLab, CapitalOne, Bangalore

Organized a 1 week training session on LLMs, their applications and current limitations, including hands-on demos.

Course Material

September 2023

Knowledge guided ECG diagnostics

At TCS Research and Innovation Lab,

I worked on improving Deep Learning based ECG diagnostic tools with the aid of domain knowledge gathered from experts. This work led to two patents - 20221052587, 202221062230 (pending).

November 2021 - February 2022

Deep Analysis of Pain Management

As a project student at ISI Kolkata

Collaborated with medical professionals in the field of radio diagnosis to formulate a problem statement and set up data gathering protocols to create a high quality data set for analysis of back pain from MRI images

August 2020 - December 2020

Adversarially robust deep learning systems

MS Thesis, under the supervision of Dr. Utpal garain.

I explored the issue of adversarial robustness using several approaches like transfer learning, cryptography, and tried to construct a mathematical model for our understanding of adversarial examples. Developed key insights into training and deploying deep learning models, experience with tensorflow, ImageNet models, and language models.

July 2019 - May 2020


Hobbies

Sports

I have played in the Senior Division Men's Basketball league of my state. I have also represented my college team in Basketball, Volleyball and Athletics.

Music

I am a classically trained pianist, and I enjoy listening to and performing works by Chopin , Beethoven, etc.

Publications

Computational Linguistics, 2023

Analyzing Semantic Faithfulness of Language Models via Input Intervention on Question Answering

Akshay Chaturvedi, Swarnadeep Bhar, Soumadeep Saha, Utpal Garain, Nicholas Asher

PLOS One, 2023

MedTric : A clinically applicable metric for evaluation of multi-label computational diagnostic systems

Soumadeep Saha, Utpal Garain, Arijit Ukil, Arpan Pal, Sundeep Khandelwal

AAAI W3PHIAI, 2024

DOST - Domain Obedient Self-supervision for Trustworthy MLC with Noisy Labels

Soumadeep Saha, Utpal Garain, Arijit Ukil, Arpan Pal, Sundeep Khandelwal


Positions and Awards

HIDA Helmholtz Visiting Researcher

2024

Visiting the German Aerospace Institute (DLR), Cologne, Germany

KVPY Fellow

2014 (SX)

Recipient of the prestigious KVPY fellowship awarded by DST, Govt. of India.

NTSE Scholar

2013

Recipient of the National Talent Search Examination scholarship.


Updates

Feb. 2024 Attending AAAI 2024, at Vancouver, BC, Canada.

Feb. 2024 New manuscript - VALUED: Vision and Logical Understanding Evaluation Dataset

Jan. 2024 New manuscript - LADDER: Revisiting the Cosmic Distance Ladder with Deep Learning Approaches and Exploring its Applications