Satyadwyoom Kumar

Greetings! I’m a Sr. Applied Scientist specializing in designing innovative algorithmic and data-driven tools which has already empowered organizations like Bain & Co., OYO Rooms, and Dell Technologies to better serve their users. Few of my notable self-engineered methods include a cosine-similarity based nearest-neighbor text search engine and a binary search based dynamic pricing system. Additionally, I have also developed scalable, production-grade NLP pipelines leveraging large language models for topic and sentiment classification, multilingual translation, and explainable behavioral assessment, enabling data-driven decision-making at scale.

My undergraduate journey, during which I delved into impactful interdisciplinary research such as in the Intelligent Data Science Lab at the University of British Columbia, Canada, collaborating with Prof. Apurva Narayan to robustify machine learning (ML) models, and in the Medical Mechatronics Lab at the National University Singapore, under Prof. Hongliang Ren, where I developed an ML solution for multimodal medical AI, has provided me with a firm theoretical groundwork to tackle real-world challenges.

These experiences have honed my technical skills and fueled my passion for enhancing the usability and trustworthiness of ML models, ensuring their seamless applicability to diverse problems. So, I invite you to delve into my works and discover how my expertise can contribute meaningfully to your projects.

PUBLICATIONS

Introducing Diversity in Feature Scatter Adversarial Training via Synthesis

Satyadwyoom Kumar, and Apurva Narayan

In 26th International Conference on Pattern Recognition (ICPR), 2022

Bib PDF

@inproceedings{kumar2022introducing,
  title = {Introducing Diversity in Feature Scatter Adversarial Training via Synthesis},
  author = {Kumar, Satyadwyoom and Narayan, Apurva},
  booktitle = {26th International Conference on Pattern Recognition (ICPR)},
  pages = {3069--3075},
  year = {2022},
  organization = {IEEE},
}

Towards Robust Certified Defense via Improved Randomized Smoothing

Satyadwyoom Kumar, and Apurva Narayan

In International Joint Conference on Neural Networks (IJCNN), 2022

Bib PDF

@inproceedings{kumar2022towards,
  title = {Towards Robust Certified Defense via Improved Randomized Smoothing},
  author = {Kumar, Satyadwyoom and Narayan, Apurva},
  booktitle = {International Joint Conference on Neural Networks (IJCNN)},
  pages = {1--8},
  year = {2022},
  organization = {IEEE},
}

Peg Transfer Workflow recognition challenge report: Do multimodal data improve recognition?

Arnaud Huaulmé, Kanako Harada, Quang-Minh Nguyen, and 10 more authors

Computer Methods and Programs in Biomedicine, 2023

Bib PDF

@article{huaulme2023peg,
  title = {Peg Transfer Workflow recognition challenge report: Do multimodal data improve recognition?},
  author = {Huaulm{\'e}, Arnaud and Harada, Kanako and Nguyen, Quang-Minh and Park, Bogyu and Hong, Seungbum and Choi, Min-Kook and Peven, Michael and Li, Yunshuang and Long, Yonghao and Dou, Qi and Kumar, Satyadwyoom and Ren, Hongliang and others},
  journal = {Computer Methods and Programs in Biomedicine},
  volume = {236},
  pages = {107561},
  year = {2023},
  publisher = {Elsevier},
}

BB-Patch: BlackBox Adversarial Patch-Attack using Zeroth-Order Optimization

Satyadwyoom Kumar, Saurabh Gupta, and Arun Balaji Buduru

arXiv, 2024

Bib PDF

@article{kumar2024bbpatch,
  title = {BB-Patch: BlackBox Adversarial Patch-Attack using Zeroth-Order Optimization},
  author = {Kumar, Satyadwyoom and Gupta, Saurabh and Buduru, Arun Balaji},
  journal = {arXiv},
  pages = {},
  year = {2024},
  organization = {arXiv},
}

GAN Inversion and Shifting: Recommending Product Modifications to Sellers for Better User Preference

Satyadwyoom Kumar, Abhijit Sharma, and Apurva Narayan

PeerJ Computer Science, 2025

Bib

@article{kumar2025gan,
  title = {GAN Inversion and Shifting: Recommending Product Modifications to Sellers for Better User Preference},
  author = {Kumar, Satyadwyoom and Sharma, Abhijit and Narayan, Apurva},
  journal = {PeerJ Computer Science},
  volume = {11},
  pages = {e2553},
  year = {2025},
  publisher = {PeerJ Inc.},
}