arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2607.06643cs.CRcs.LG
收藏

社区训练中后门吸收的力量

The Power of Backdoor Absorption in Community Training

Université Paris-Saclay, CEA LIST(巴黎萨克雷大学,CEA LIST)

Issam Seddik, Sami Souihi, Mohamed Tamaazousti, Sara Tucci Piergiovanni

AI总结:

研究在分散训练中后门攻击问题,通过量化最小审计开销,将注入-吸收动态形式化为离散时间马尔可夫链,证明结合自然吸收等策略可使对手成功概率降为零,实证显示能有效抑制后门且效用无退化,为安全关键型AI提供高效防御。

摘要:

后门攻击严重威胁大规模人工智能模型。在分散训练范式中,当模型所有者将训练委托给外部计算提供商时,对手可以精心设计隐蔽的低频触发器来注入恶意行为,同时规避标准审计。传统上,检测这些攻击需要完全重新计算训练步骤,这一开销过高,与所有者的资源限制直接矛盾。为解决此问题,我们研究了拜占庭扰动下连续优化动态的弹性,其中对手被迫与持续涌入的诚实更新竞争。在对手破坏n个训练者中的f个的威胁模型下,我们量化了模型所有者为概率性地限制攻击成功率所需的最小审计开销。我们将这种注入-吸收动态形式化为离散时间马尔可夫链(DTMC)。使用此框架,我们证明在结合自然吸收、随机调度器和惰性验证预言机的防御策略下,任何有界对手的成功概率渐近地降至零。实证结果表明,即使仅在10%的总训练步骤上调用验证预言机,也能显著抑制后门且效用无退化。这种方法为安全关键型人工智能提供了一种可证明合理且计算高效的防御。

英文摘要:

Backdoor attacks severely threaten large-scale AI models. When model owners delegate training to external compute providers within a decentralized training paradigm, adversaries can craft stealthy, low-frequency triggers to inject malicious behavior while evading standard audits. Traditionally, detecting these attacks requires a full re-computation of the training steps--a prohibitive overhead that directly contradicts the owner's resource constraints. To address this, we investigate the resilience of continuous optimization dynamics under Byzantine perturbations, where adversaries are forced to compete against a continuous influx of honest updates. Under a threat model where an adversary compromises f out of n total trainers, we quantify the minimum auditing overhead required by the model owner to probabilistically bound the attack success rate. We formalize this injection-absorption dynamic as a Discrete-Time Markov Chain (DTMC). Using this framework, we prove that the success probability of any bounded adversary asymptotically collapses to zero under a defense strategy combining natural absorption, a randomized scheduler, and lazy verification oracle. Empirical results demonstrate significant backdoor suppression with zero utility degradation even when invoking the verification oracle on merely 10% of the total training steps. This approach yields a provably sound and computationally efficient defense for safety-critical AI.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑