Two of my papers were accepted to European Alliance for Innovation - EAI's Mobiquitous 2020 (CORE A). One paper "Quantifying Privacy Leakage in Graph Embeddings" addresses privacy risks in Graph learning models. The second titled "Towards enhancing Fault Tolerance of NNs" improves the reliability of NNs.

Our recent work showing model extraction attacks are realistic in image to image translation models based on GANs

Paper titled "Inferring Sensitive Attributes from Model Explanations" has been accepted to CIKM 2022. We show a tradeoff between privacy and transparency in ML. We show that model explanations can enable an adversary to infer sensitive attributes. Paper: arxiv.org/pdf/2208.09967…

Ever found the discourse around "intelligence" in "A(G)I" squicky or heard folks pointing out the connection w/eugenics & wondered what that was about? History of it all can be found in this excellent talk by @timnitGebru (@dair-community.social/bsky.social) (w/ co-author Dr. Émile P. Torres) youtube.com/watch?v=P7XT4T…

Interesting discussions on catastrophic risks of AI with Yoshua Bengio and Philippe Beaudoin

Our paper "GrOVe: Ownership Verification of Graph Neural Networks using Embeddings" has been accepted to IEEE S&P. We propose the first fingerprinting scheme against model extraction attacks in GNNs. w/ Asim Waheed and N. Asokan [arxiv.org/abs/2304.08566]

New paper to appear at ESORICS’24: "Attesting Distributional Properties of Training Data for Machine Learning" by Vasisht Duddu, Anudeep Das, Nora Khayata, @HYalame, Thomas Schneider and N. Asokan. Full version: arxiv.org/abs/2308.09552

Our new paper shows how to prove in a privacy-preserving way that an ML model was properly trained on proper data that fulfills certain properties. Great collaboration between N. Asokan's Secure Systems Group CrySP University of Waterloo @UWaterloo and ENCRYPTO Computer Science, TU Darmstadt TU Darmstadt

We also introduce the notion of "property attestation" (adapted from two-decade-old trusted/confidential computing research!) for ML models, useful for demonstrating #regulatory #compliance of #ArtificialIntelligence models Vasisht Duddu Thomas Schneider Nora Khayata @HYalame

The research community has been exploring various risks in #ML-based systems and developing defenses. But when a defense against a specific risk is deployed, how does it affect _other_ risks? We explain more in this blog article blog.ssg.aalto.fi/2024/05/uninte… Secure Systems Aalto CrySP @UWaterloo

Thank you. I'm grateful to both of you for your guidance, and the program committee for recognizing our work with an award.

It is gratifying when universities take care to celebrate the achievements of their students: Waterloo's Cheriton School of Computer Science highlights a recent achievement of doctoral student Vasisht Duddu and former research visitor @sebszyller. cs.uwaterloo.ca/news/vasisht-d…

🎉Vasisht Duddu and Rasoul Akhavan Mahdavi have won the Cybersecurity & Privacy Excellence Graduate Scholarships. 🤖This $10K award will support their research in machine learning and privacy-preserving technologies. 🔗Read more: uwaterloo.ca/cybersecurity-… #UWaterloo

Congratulations to Vasisht Duddu on being awarded a prestigious IBM PhD Fellowship. One of only 24 fellowship recipients worldwide, Vasisht’s research enhances trust in machine learning systems. Learn more about his work and its impact: cs.uwaterloo.ca/news/vasisht-d…

Honored to receive the PhD fellowship from IBM Research. I am deeply grateful to my advisor, Prof. N. Asokan, for his constant support and mentorship, and all my wonderful collaborators and colleagues.

My student Anudeep Das initiated this work when he was an undergrad. Working relentlessly with PhD students Vasisht Duddu & Rui Zhang, has won an ACM CODASPY '26 Best Paper Award. This is rare, even in a place like University of Waterloo with high-achieving undergrads! cs.uwaterloo.ca/news/anudeep-d…