Debaditya Shome

Debaditya Shome

Graduate researcher experienced in AI / ML

Queen's University

Biography

I’m a graduate student pursuing MASc in AI (research) at Queen’s University. My research interests include Generative models, Self-supervised learning, Multi-modality, and learning universal representations of time-series. I’m an author of 10+ publications, encompassing a total of 100+ citations. Apart from research, I have prior experience in Data engineering and deploying Machine learning models at scale on cloud/serverless to integrate with real-world products/applications.

Achievements:

  • Vector Scholarship in AI
  • MITACS Globalink Graduate Fellow
Education
  • MASc in Artificial Intelligence, September 2022 to Present

    Queen's University

  • B.Tech in Electronics and Telecommunication, July 2018 to July 2022

    KIIT University

Experience

 
 
 
 
 
Ingenuity Labs & AIIM lab
Graduate research assistant
Sep 2022 – Present Kingston, Canada

Overview:

  • Graduate research supervised by Dr. Ali Etemad.
  • Implemented a state-of-the-art High-Fidelity PPG-to-ECG translation system powered by a novel class of Diffusion Models. Demonstrated the ability to detect a range of Cardiac conditions/diseases using synthetic ECGs with significantly higher F1 than the input PPGs. Paper under review at an A* conference.
  • Developed a Speech Emotion classifier capable of explicitly understanding the linguistic and prosodic aspect of emotions using Cross-modal Knowledge distillation. Experiments show state-of-the-art performance on IEMOCAP. Paper to be submitted at an A* conference.
 
 
 
 
 
SHL
AI intern (Research)
Oct 2021 – Apr 2022 Gurgaon, India

Overview:

  • Presently working on few-shot prompt learning on spoken content, for SHL’s Interview Intelligence platform.
  • Developed algorithms for repititive phrase, filler phrase, self introduction and organization introduction detection.
 
 
 
 
 
Xu Lab, Carnegie Mellon University
Research intern
Sep 2021 – Nov 2021 Pennsylvania, USA

Overview:

  • Collaborated with a PhD student for a project focused on modeling continuous conformational changes in cryo-ET images with Unsupervised representation learning under the supervision of Dr. Min Xu.
  • Conducted a comprehensive literature review and baseline method implementations.
 
 
 
 
 
MITACS
MITACS Globalink Research intern
May 2021 – Aug 2021 British Columbia, Canada

Overview:

  • Conducted cross-disciplinary research at the intersection of Deep learning and wireless communications, under the supervision of Dr. Omer Waqar from Thompson River’s University, Canada.
  • Authored a comprehensive review paper addressing a gap in literature on the bi-directional interplay of Federated learning and wireless communications, accepted at the journal - Transactions on Emerging Telecommunications Technologies.
  • Designed a novel unsupervised learning algorithm for energy and power optimization in UAV networks. The paper was presented at IEEE UEMCON 2021, and recieved the Best Presenter award.
 
 
 
 
 
Samsung Research
Research intern (PRISM program)
Apr 2021 – Oct 2021 Bangalore, India

Overview:

  • Leading the project on variable-length synthetic handwriting image generation using Generative Adversarial networks.
  • Academia-Industry colloraboration with a 6-member team consisting of myself, another student, Prof. Vimal Srivastava, Prof. Manoranjan Kumar and two mentors from Samsung Research, Bangalore.
  • Generated synthetic data is being utilized to train better handwritten text recognition (HTR) models for HTR feature in Samsung smartphone’s OCR system.
 
 
 
 
 
Brand Love Intelligence
Lead Data Scientist
Apr 2021 – Nov 2021 Assen, Netherlands

Overview:

  • Focused on building full stack data science pipeline from data collection to model deployment for powering the AI engine of Relevense (https://www.relevense.com/), a Flagship market intelligence product co-funded with grants of the Europees Fonds voor Regionale Ontwikkeling (EFRO) and Samenwerkingsverband Noord Nederland (SNN).
  • Projects include Tweet based emotion recognition API, Big-5 personality classication API, Facial expression recognition, Receptive audience recommendation system.
  • Reduced the AWS costs by 60% by shifting the backend to a serverless architecture with multiple Lambda functions, DynamoDB, Timestream and S3.
  • Single handedly created end-to-end ML pipelines with all models beyond 95% accuracy along with efficient monitoring of out of training distribution inference events.
 
 
 
 
 
Omdena
Lead Machine learning Engineer
Sep 2020 – Apr 2021 California, USA

Overview:

  • Collaborated with a team of 48 while working with our client, World Resources Institute (https://www.wri.org/) on a project leveraging NLP to find geographical locations with climate hazards and potential gaps for minimizing climate change impacts across the globe. Deployed a dashboard designed with Streamlit for easy inference. Technical blog on the project: https://omdena.com/blog/climate-change-impacts/ .
  • Led a team of 48 for building the data processing and machine learning backend for a Dutch client’s market intelligence product. Got a full-time offer from the client due to extraordinary contributions in the project.
 
 
 
 
 
Emory University
Research intern
Feb 2021 – May 2021 Atlanta, USA

Overview:

  • Built an end-to-end NLP pipeline for Multi-document abstractive summarization of Radiology reports of COVID-19 patients
  • Trained longformer and BERT models on a Slurm multi-GPU cluster in an HIPAA protected server.
  • Achieved a ROUGE-1 score of 0.410 on test dataset.
 
 
 
 
 
HighRadius
Product and Engineering intern
Jan 2021 – Mar 2021 Odisha, India

Overview:

  • Developed a full stack web-based invoice management application following an end-to-end Data science product development lifecycle guided by mentors.
  • Responsibilities included identifying appropriate user requirements, designing a great user experience and building appropriate data pipelines and machine learning models along with relevant UI components and backend design.
  • Developed a state-of-the-art payment prediction system using XGboost regression, with a root-mean-squared error of 0.1 on 5-fold cross validation.

Recent Publications

Quickly discover relevant content by filtering publications.

Certifications

Computer Vision nanodegree
See certificate
edX
6.86x: Machine Learning with Python-From Linear Models to Deep Learning
See certificate
edX
18.6501x: Fundamentals of Statistics
See certificate
Coursera
Introduction to Applied Machine Learning
See certificate
Coursera
DeepLearning.AI TensorFlow Developer Specialization
See certificate
Coursera
Deep learning specialization
See certificate
Coursera
An Introduction to Practical Deep Learning
See certificate
Coursera
Operating Systems and You; Becoming a Power User
See certificate
Coursera
IBM Data Science Specialization
See certificate

Projects

Coming soon, website under construction !

Recent & Upcoming Talks

Coming soon, website under construction !

Recent Posts

Coming soon, blog under construction!

Contact