arXiv subjects

Ching-Chia Kao

Publications and source records attributed to Ching-Chia Kao.

5 recordsLinked to original sources

arXiv2025

The Great Contradiction Showdown: How Jailbreak and Stealth Wrestle in Vision-Language Models?

Vision-Language Models (VLMs) have achieved remarkable performance on a variety of tasks, yet they remain vulnerable to jailbreak attacks that compromise safety and reliability. In this paper, we provide an information-theoretic framework for understanding the fundamental trade-off between the effectiveness of these attacks and their stealthiness. Drawing on Fano's inequality, we demonstrate how an attacker's success probability is intrinsically linked to the stealthiness of generated prompts. Building on this, we propose an efficient algorithm for detecting non-stealthy jailbreak attacks, offering significant improvements in model robustness. Experimental results highlight the tension between strong attacks and their detectability, providing insights into both adversarial strategies and defense mechanisms.

cs.LG↗

arXiv2025

Safety Alignment Depth in Large Language Models: A Markov Chain Perspective

Large Language Models (LLMs) are increasingly adopted in high-stakes scenarios, yet their safety mechanisms often remain fragile. Simple jailbreak prompts or even benign fine-tuning can bypass these protocols, underscoring the need to understand where and how they fail. Recent findings suggest that vulnerabilities emerge when alignment is confined to only the initial output tokens. Unfortunately, even with the introduction of deep safety alignment, determining the optimal safety depth remains an unresolved challenge. By leveraging the equivalence between autoregressive language models and Markov chains, this paper offers the first theoretical result on how to identify the ideal depth for safety alignment, and demonstrates how permutation-based data augmentation can tighten these bounds. Crucially, we reveal a fundamental interaction between alignment depth and ensemble width-indicating that broader ensembles can compensate for shallower alignments. These insights provide a theoretical foundation for designing more robust, scalable safety strategies that complement existing alignment approaches, opening new avenues for research into safer, more reliable LLMs.

cs.LG↗

arXiv2024

Defending Against Repetitive Backdoor Attacks on Semi-supervised Learning through Lens of Rate-Distortion-Perception Trade-off

Semi-supervised learning (SSL) has achieved remarkable performance with a small fraction of labeled data by leveraging vast amounts of unlabeled data from the Internet. However, this large pool of untrusted data is extremely vulnerable to data poisoning, leading to potential backdoor attacks. Current backdoor defenses are not yet effective against such a vulnerability in SSL. In this study, we propose a novel method, Unlabeled Data Purification (UPure), to disrupt the association between trigger patterns and target classes by introducing perturbations in the frequency domain. By leveraging the Rate-Distortion-Perception (RDP) trade-off, we further identify the frequency band, where the perturbations are added, and justify this selection. Notably, UPure purifies poisoned unlabeled data without the need of extra clean labeled data. Extensive experiments on four benchmark datasets and five SSL algorithms demonstrate that UPure effectively reduces the attack success rate from 99.78% to 0% while maintaining model accuracy. Code is available here: \url{https://github.com/chengyi-chris/UPure}.

cs.CV↗

arXiv2024

Minimal Cascade Gradient Smoothing for Fast Transferable Preemptive Adversarial Defense

Adversarial attacks persist as a major challenge in deep learning. While training- and test-time defenses are well-studied, they often reduce clean accuracy, incur high cost, or fail under adaptive threats. In contrast, preemptive defenses, which perturb media before release, offer a practical alternative but remain slow, model-coupled, and brittle. We propose the Minimal Sufficient Preemptive Defense (MSPD), a fast, transferable framework that defends against future attacks without access to the target model or gradients. MSPD is driven by Minimal Cascade Gradient Smoothing (MCGS), a two-epoch optimization paradigm executed on a surrogate backbone. This defines a minimal yet effective regime for robust generalization across unseen models and attacks. MSPD runs at 0.02s/image (CIFAR-10) and 0.26s/image (ImageNet), 28--1696 times faster than prior preemptive methods, while improving robust accuracy by +5% and clean accuracy by +3.7% across 11 models and 7 attacks. To evaluate adaptive robustness, we introduce Preemptive Reversion, the first white-box diagnostic attack that cancels preemptive perturbations under full gradient access. Even in this setting, MSPD retains a +2.2% robustness margin over the baseline. In practice, when gradients are unavailable, MSPD remains reliable and efficient. MSPD, MCGS, and Preemptive Reversion are each supported by formal theoretical proofs. The implementation is available at https://github.com/azrealwang/MSPD.

cs.CR↗

arXiv2020

Deterministic Certification to Adversarial Attacks via Bernstein Polynomial Approximation

Randomized smoothing has established state-of-the-art provable robustness against $\ell_2$ norm adversarial attacks with high probability. However, the introduced Gaussian data augmentation causes a severe decrease in natural accuracy. We come up with a question, "Is it possible to construct a smoothed classifier without randomization while maintaining natural accuracy?". We find the answer is definitely yes. We study how to transform any classifier into a certified robust classifier based on a popular and elegant mathematical tool, Bernstein polynomial. Our method provides a deterministic algorithm for decision boundary smoothing. We also introduce a distinctive approach of norm-independent certified robustness via numerical solutions of nonlinear systems of equations. Theoretical analyses and experimental results indicate that our method is promising for classifier smoothing and robustness certification.

cs.LG↗