Aniruddha Saha

Applied Scientist Intern at AWS AI


This summer, I'll be an Applied Scientist Intern at Amazon Rekognition and Video.

I am currently pursuing my PhD in Computer Science at University of Maryland Baltimore County advised by Hamed Pirsiavash. I am broadly interested in identifying and mitigating adversarial vulnerabilities in deep learning models.

My research has involved studying Backdoor Attacks in CNNs - curating attacks and building fast solutions to identify backdoors. I have also investigated how spatial context introduces vulnerabilities in fast object detectors like YOLO and how to train detectors robust to contextual adversarial patches. I am currently studying unsupervised and self-supervised computer vision algorithms for application in computer-assisted colonoscopy procedures.

Summer of 2019, I was an intern at Matroid where I developed Computer Vision solutions for Matroid studio. I was advised by Reza Zadeh and Danny Jeck. I implemented OpenPose by CMU to run end-to-end inference on Tensorflow GPU for 2D human body pose estimation. It's now featured on Matroid Studio. Try it out here: Matroid Public Detectors

Prior to this, I was a Software Engineer at Samsung Research Institute Bangalore, India where I was part of the DRAM Group in the Memory Team of Samsung Semiconductor India Research.

I spend my time in photography and writing among other things. I am a big football fan and I support Manchester United FC.


Most recent publications on Google Scholar.

Role of Spatial Context in Adversarial Robustness for Object Detection Paper Slides Video Code

Aniruddha Saha*, Akshayvarun Subramanya*, Koninika Patil, Hamed Pirsiavash

CVPR 2020 Workshop on Adversarial Machine Learning in Computer Vision

*equal contribution

Universal Litmus Patterns: Revealing Backdoor Attacks in CNNs Paper Webpage Slides Video Code

Soheil Kolouri*, Aniruddha Saha*, Hamed Pirsiavash, Heiko Hoffmann

CVPR 2020 Oral

*equal contribution

Hidden Trigger Backdoor Attacks Paper Slides Poster Code

Aniruddha Saha, Akshayvarun Subramanya, Hamed Pirsiavash

AAAI 2020 Oral

An Adaptive Foreground-Background Separation Method for Effective Binarization of Document Images Paper

Bishwadeep Das, Showmik Bhowmik, Aniruddha Saha, Ram Sarkar

Proceedings of the Eighth International Conference on Soft Computing and Pattern Recognition (SoCPaR 2016)

Academic Service - Reviewer

This website uses the website design and template by Martin Saveski.