Abstract:Deep learning technology has brought convenience and advanced developments but has become untrustworthy due to its sensitivity to adversarial attacks. Attackers may utilize this sensitivity to manipulate predictions. To defend against such attacks, existing anti-adversarial methods typically counteract adversarial perturbations post-attack, while we have devised a proactive strategy that preempts by safeguarding media upfront, effectively neutralizing potential adversarial effects before the third-party attacks occur. This strategy, dubbed Fast Preemption, provides an efficient transferable preemptive defense by using different models for labeling inputs and learning crucial features. A forward-backward cascade learning algorithm is used to compute protective perturbations, starting with forward propagation optimization to achieve rapid convergence, followed by iterative backward propagation learning to alleviate overfitting. This strategy offers state-of-the-art transferability and protection across various systems. With the running of only three steps, our Fast Preemption framework outperforms benchmark training-time, test-time, and preemptive adversarial defenses. We have also devised the first, to our knowledge, effective white-box adaptive reversion attack and demonstrate that the protection added by our defense strategy is irreversible unless the backbone model, algorithm, and settings are fully compromised. This work provides a new direction to developing proactive defenses against adversarial attacks.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the problem of the sensitivity of deep - learning models to adversarial attacks. Specifically, the author proposes a new active defense strategy - Fast Preemption - to enhance the security of input data in advance, thereby effectively neutralizing potential adversarial effects before third - party attacks occur. #### Background problems 1. **The threat of adversarial attacks**: Although deep - learning technology has brought convenience and advanced development, its sensitivity to adversarial attacks makes it untrustworthy. Attackers can manipulate prediction results by taking advantage of this sensitivity, for example, by changing the content of images, audio or video through tiny perturbations. 2. **Limitations of existing defense methods**: - **Passive defense**: It usually counteracts after an attack occurs, which may reduce classification accuracy and is vulnerable to counter - attacks. - **Active defense**: Although it can protect before an attack, it is inefficient in real - time applications and usually requires paired modules, resulting in the protection not being portable to other systems. - **Defense during training and testing**: The former requires a large amount of data and may reduce the clean accuracy rate; the latter is difficult to detect unknown attacks and is prone to introducing errors. #### Solutions To overcome these problems, the author proposes the Fast Preemption strategy, which has the following characteristics: 1. **Pre - defense**: Fast Preemption "attacks" the media on the user side in advance, that is, enhances it before third - party attacks, to neutralize potential adversarial effects. 2. **Efficient and portable**: Different models are used for input tagging and feature learning, and the Fore - Back Cascade Learning Algorithm is adopted to calculate the protective perturbation. This method not only improves the convergence speed but also reduces over - fitting and provides better portability. 3. **No need for real labels**: There is no need to manually label each input, and an automatic classifier can perform this task. 4. **Fast execution**: The defense can be completed in just three steps, significantly reducing the time cost. 5. **Irreversible protection**: The first effective white - box adaptive reversion attack (Preemptive Reversion) is designed, proving that the protection of Fast Preemption is irreversible unless the backbone model, algorithm and settings are completely leaked. ### Summary Fast Preemption provides a new active defense strategy that can more effectively resist adversarial attacks while maintaining high clean accuracy and robust accuracy, and has efficient portability. In addition, the author also proposes a comprehensive evaluation protocol for evaluating the reliability of pre - processing defenses. --- If you have more questions or need further explanation, please feel free to let me know!

Fast Preemption: Forward-Backward Cascade Learning for Efficient and Transferable Proactive Adversarial Defense

Attack As Defense: Characterizing Adversarial Examples Using Robustness.

A Universal Defense Strategy Against Adversarial Attacks Based on Attention-Guided

An Efficient Preprocessing-based Approach to Mitigate Advanced Adversarial Attacks

Test-time Adversarial Defense with Opposite Adversarial Path and High Attack Time Cost

Adversarial Feature Map Pruning for Backdoor

Towards Understanding Fast Adversarial Training

Embodied Active Defense: Leveraging Recurrent Feedback to Counter Adversarial Patches

DeepFense: Online Accelerated Defense Against Adversarial Deep Learning

Mitigating Backdoor Attack by Injecting Proactive Defensive Backdoor

You Only Propagate Once: Painless Adversarial Training Using Maximal Principle

Learn from the Past: A Proxy Guided Adversarial Defense Framework with Self Distillation Regularization

Protecting Feed-Forward Networks from Adversarial Attacks Using Predictive Coding

You Only Propagate Once: Accelerating Adversarial Training Via Maximal Principle

Beating Backdoor Attack at Its Own Game

PubDef: Defending Against Transfer Attacks From Public Models

A Framework for Robust Deep Learning Models Against Adversarial Attacks Based on a Protection Layer Approach

Efficient Two-Step Adversarial Defense for Deep Neural Networks

Towards Deep Learning Models Resistant to Transfer-based Adversarial Attacks via Data-centric Robust Learning

Deflecting Adversarial Attacks

How to Steer Your Adversary: Targeted and Efficient Model Stealing Defenses with Gradient Redirection