Toward Efficient Robust Training against Union of ℓ_p Threat Models
Nov 28, 2022
Sprecher:innen
Gaurang Sriramanan
Řečník · 1 sledující
Maharshi Gor
Řečník · 1 sledující
Soheil Feizi
Řečník · 2 sledující
Über
The overwhelming vulnerability of deep neural networks to carefully crafted perturbations known as adversarial attacks has led to the development of various training techniques to produce robust models. While the primary focus of existing approaches has been directed toward addressing the worst-case performance achieved under a single-threat model, it is imperative that safety-critical systems are robust with respect to multiple threat models simultaneously. Existing approaches that address worst-case performance under the union of such threat models (ℓ_∞, ℓ_2, ℓ_1) either utilize adversarial training methods that require multi-step attacks which are computationally expensive in practice, or rely upon fine-tuning of pre-trained models that are robust with respect to a single-threat model. In this work, we show that by carefully choosing the objective function used for robust training, it is possible to achieve similar, or improved worst-case performance over a union of threat models while utilizing only single-step attacks, thereby achieving a significant reduction in computational resources necessary for training. Furthermore, prior work showed that adversarial training specific to the ℓ_1 threat model is relatively difficult, to the extent that even multi-step adversarially trained models were shown to be prone to gradient-masking. However, the proposed method—when applied on the ℓ_1 threat model specifically—enables us to obtain the first ℓ_1 robust model trained solely with single-step adversaries. Finally, to demonstrate the merits of our approach, we utilize a modern set of attack evaluations to better estimate the worst-case performance under the considered union of threat models.The overwhelming vulnerability of deep neural networks to carefully crafted perturbations known as adversarial attacks has led to the development of various training techniques to produce robust models. While the primary focus of existing approaches has been directed toward addressing the worst-case performance achieved under a single-threat model, it is imperative that safety-critical systems are robust with respect to multiple threat models simultaneously. Existing approaches that address wors…
Organisator
NeurIPS 2022
Účet · 960 sledujících
Gefällt euch das Format? Vertraut auf SlidesLive, um euer nächstes Event festzuhalten!
Professionelle Aufzeichnung und Livestreaming – weltweit.
Freigeben
Empfohlene Videos
Präsentationen, deren Thema, Kategorie oder Sprecher:in ähnlich sind
LASSIE: Learning Articulated Shapes from Sparse Image Ensemble via 3D Part Discovery
Zhlédnout později
Chun-Han Yao, …
Pro uložení prezentace do věčného trezoru hlasovalo 0 diváků, což je 0.0 %
ConvMAE: Masked Convolution Meets Masked Autoencoders
Zhlédnout později
Peng Gao, …
Pro uložení prezentace do věčného trezoru hlasovalo 0 diváků, což je 0.0 %
What Makes a Good Explanation?: A Unified View of Properties of Interpretable ML
Zhlédnout později
Pro uložení prezentace do věčného trezoru hlasovalo 0 diváků, což je 0.0 %
LILA: A Unified Benchmark for Mathematical Reasoning
Zhlédnout později
Pro uložení prezentace do věčného trezoru hlasovalo 0 diváků, což je 0.0 %
Active Learning with Safety Constraints
Zhlédnout později
Pro uložení prezentace do věčného trezoru hlasovalo 0 diváků, což je 0.0 %
Robust Learning against Relational Adversaries
Zhlédnout později
Yizhen Wang, …
Pro uložení prezentace do věčného trezoru hlasovalo 0 diváků, což je 0.0 %