Ankita Raj

I am a PhD candidate in the Computer Science Department at IIT Delhi. I am working on robust and trustworthy algorithms for Machine Learning, specifically in the context of Computer Vision. I am advised by Dr. Chetan Arora.

I have a Master's degree in Electronics and Communication Engineering from IIIT Delhi, where I was advised by Dr. Pravesh Biyani. I also held a research position in the Robotics lab, Tata Consultancy Services from 2016 to 2018, where I worked on computer vision for robotics.

Email / CV / Google Scholar / Github / LinkedIn

Research

My research is aimed at developing robust computer-vision systems. I am working on techniques for mounting and preventing various types of attacks on learning-based computer vision systems, including backdoor attacks and model-stealing attacks.

In the past, I have worked on incremental learning and object detection for computer vision, and optimization for resource allocation.

	Examining the Threat Landscape: Foundation Models and Model Theft Ankita Raj, Deepankar Varma , Chetan Arora British Machine Vision Conference (BMVC), 2024 pdf · Supplementary · Video · Poster We show that image classification models adapted from foundation models (such as ViTs) on downstream tasks are more vulnerable to model stealing attacks compared to models derived from conventional vision architectures like ResNets.
	Assessing Risk of Stealing Proprietary Models for Medical Imaging Tasks Ankita Raj, Harsh Swaika , Deepankar Varma , Chetan Arora International Conference on Medical Imaging and Computer Assisted Intervention (MICCAI ), 2024 Springer link · pdf · Supplementary · Poster · Code We explore the vulnerability of proprietary medical imaging models to model stealing attacks and propose a novel attack method that efficiently steals models under realistic threat scenarios.
	Identifying Physically Realizable Triggers for Backdoored Face Recognition Networks Ankita Raj, Ambar Pal, Chetan Arora International Conference on Image Processing, 2021 IEEE Xplore · pdf · Video · Poster Given a face recognition network potentially backdoored with physically realizable triggers like sunglasses or bowtie, we propose a technique to identify such triggers.
	Weighted-A* Based Energy Efficient Resource Allocation in G. Fast Ankita Raj, Pravesh Biyani IEEE Transactions on Communications, 2019 IEEE Xplore · pdf We formulate and solve an optimization problem to minimize power consumption in G.fast broadband access technology.
	HiFI: A Hierarchical Framework for Incremental Learning using Deep Feature Representation Ankita Raj, Anima Majumder, Swagat Kumar IEEE International Conference on Robot and Human Interactive Communication (RO-MAN), 2019 IEEE Xplore · pdf A hierarchical deep-learning framework for object recognition, with near real-time training for recognizing previously unseen objects.
	A* algorithm based power minimization for discontinuous operations in G. fast Ankita Raj, Pravesh Biyani, Sandip Aine IEEE International Conference on Communications (ICC), 2017 IEEE Xplore · pdf We schedule users to time slots during discontinuous operation mode in G.fast, to achieve high energy efficiency.
	Analysis and Synthesis Prior Greedy Algorithms for Non-linear Sparse Recovery Kavya Gupta, Ankita Raj, Angshul Majumdar Data Compression Conference (DCC), 2016 IEEE Xplore · arxiv · pdf We propose algorithms for recovering sparse solutions to non-linear inverse problems.

Recent Updates

[December 2024] Attended ICVGIP 2024 at IIIT Bangalore.
[July 2024] Paper accepted at BMVC 2024 (Oral).
[June 2024] Paper accepted at MICCAI 2024.
[March 2024] Winner of "3 Minute Thesis" competition organized by Research Scholar Forum, IIT Delhi. Here is my slide for the competition.
[September 2023] Delivered a hands-on session on "Vision Transformers for Brain Tumor Classification in MRI Images" in the workshop on “Artificial Intelligence in Modern Biology” sponsored by DBT at ICGEB, New Delhi.

Teaching

Teaching Assistant (at IIT Delhi): Machine Learning, Digital Image Processing, Computer Vision, Datastructures and Algorithms, Introduction to Computer Science, Advanced Functional Brain Imaging.
Teaching Assistant (at IIIT Delhi): Probability and Statistics, Engineering Optimization, Signals and Systems, Maths I (Linear Algebra), Maths II (Differential Equations).

Services

Reviewer / PC Member for conferences: CVPR (2025, 2024), WACV (2025, 2024, 2023), AAAI (2026, 2025), MICCAI (2025, 2024), WIFS 2025, ICCV SafeMM Workshop 2025.
Reviewer for journals: International Journal of Computer Vision, IEEE Transactions on Multimedia.

Template: Jon Barron's awesome website