aiqbench analysis
This research proposes a novel method to accelerate adversarial example generation by predicting input gradients from forward-pass hidden states, effectively eliminating the computationally expensive backward pass. By leveraging lightweight linear regression and theoretical derivations within the Neural Tangent Kernel regime, the approach achieves a 532% increase in throughput while maintaining attack performance comparable to standard FGSM methods. This advancement directly addresses scalability bottlenecks in robustness evaluation and red-teaming workflows for machine learning models.
Fast Adversarial Attacks with Gradient Prediction
arXiv:2605.14868v2 Announce Type: replace Abstract: Generating adversarial examples at scale is a core primitive for robustness evaluation, adversarial training, and red-teaming, yet even "fast" attacks such as FGSM remain throughput-limited by the cost of a backward pass. We introduce a family of attacks that eliminates the backward pass by predicting the input gradient from forward-pass hidden states via a lightweight linear regression. Theoretically, we derive exact affine conditional gradient means, showing optimality in the (idealized) NTK regime. Empirically, our methods work when applied to practical finite-width models; we recover much of FGSM's attack performance while using only a small fraction of the time, corresponding to a $532\%$ increase in throughput. These results suggest gradient prediction as a simple and general route to significantly faster adversarial generation under realistic wall-clock constraints.
Abstract:Generating adversarial examples at scale is a core primitive for robustness evaluation, adversarial training, and red-teaming, yet even "fast" attacks such as FGSM remain throughput-limited by the cost of a backward pass. We introduce a family of attacks that eliminates the backward pass by predicting the input gradient from forward-pass hidden states via a lightweight linear regression. Theoretically, we derive exact affine conditional gradient means, showing optimality in the (idealized) NTK regime. Empirically, our methods work when applied to practical finite-width models; we recover much of FGSM's attack performance while using only a small fraction of the time, corresponding to a $532\%$ increase in throughput. These results suggest gradient prediction as a simple and general route to significantly faster adversarial generation under realistic wall-clock constraints.
Machine Learning (cs.LG)
arXiv:2605.14868 [cs.LG]
(or arXiv:2605.14868v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2605.14868 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 217 words
Related coverage
A Robust and Explainable Transformer-Based Framework for Phishing Email Detection
arXiv · Jun 3, 2026
Problems of Proliferation
LessWrong · Sep 29, 2026
Mitigating Black-Box Score-Based Query Attacks via Gradient Confusion
arXiv · Oct 15, 2026
Information-Geometric Inverse Distillation for Enhancing Adversarial Transferability
arXiv · Sep 18, 2026