The Devil’s Advocate: Shattering the Illusion of Unexploitable Data using Diffusion Models
Published in Preprint, 2023
In this work, we reveal the fragility of unlearnable examples using diffusion models.
Publications
Adversarial Coreset Selection for Efficient Robust Training
Published in Preprint, 2022
In this paper, we theoretically analyze adversarial coreset selection and provide insights into its working dynamics.
COLLIDER: A Robust Training Framework for Backdoor Data
Published in The 16th Asian Conference on Computer Vision (ACCV), 2022
In this paper, we utilize adaptive subset selection to eliminate backdoor data and train a clean model.
ℓ∞-Robustness and Beyond: Unleashing Efficient Adversarial Training
Published in The 17th European Conference on Computer Vision (ECCV), 2022
In this work, we reduce the training time of adversarial training using adaptive sample selection.
AdvFlow: Inconspicuous Black-box Adversarial Attacks using Normalizing Flows
Published in The 34th Conference on Neural Information Processing Systems (NeurIPS), 2020
In this paper, we propose a novel black-box adversarial attack that exploits clean data distribution to conceal itself.
Invertible Generative Modeling using Linear Rational Splines
Published in The 23rd International Conference on Artifcial Intelligence and Statistics (AISTATS), 2020
In this paper, we explore using linear rational splines as an invertible transformations for normalizing flows.
Deterministic Design of Toeplitz Matrices With Small Coherence Based on Weyl Sums
Published in IEEE Signal Processing Letters, 2019
This paper is about designing Toeplitz measurement matrices based on Weyl sums.
A Sampling Theorem for Convex Shapes with Algebraic Boundaries
Published in International Conference on Sampling Theory and Applications (SampTA), 2017
In this paper we present a sampling result about continuous-domain black and white images that form a convex shape.