M3C:通过缓解先前一致的阶级混乱来抵御不可知论攻击
IEEE transactions on pattern analysis and machine intelligence
|September 25, 2025
概括
新的对抗性训练 (AT) 方法通过减轻一致的类混 (3C) 来抵御未知的攻击. 这种方法增强了深度神经网络 (DNN) 对各种对抗性干扰的概括性.
科学领域:
- 人工智能的人工智能
- 机器学习 机器学习
- 计算机视觉 计算机视觉
背景情况:
- 深度神经网络 (DNN) 面临着来自对抗性攻击的重大安全挑战.
- 敌对训练 (AT) 是有效的,但往往缺乏针对隐形攻击的概括性.
- 由于有限的稳定性,现有的AT方法与训练不可知攻击作斗争.
研究的目的:
- 为了解决当前对抗训练 (AT) 方法的有限通用性.
- 提出一个统一的防御战略,以对抗广泛的敌对攻击,包括训练不可知的攻击.
- 引入一种新的方法,利用一贯的类混 (3C) 提高强度.
主要方法:
- 在不同攻击中,在AT分类器中发现了一个可概括的先前:一致的类混 (3C).
- 提出了一种缓解一致类混 (M3C) 模型,以提高AT的概括性.
- 优化了不确定性加权的对抗性混损失 (ACL),以专注于混的样本.
- 引入了渐变意识注意力 (GAA) 机制,以抑制恶性特征并增强正确的类信心.
主要成果:
- M3C模型显著提高了AT强度对抗不可知攻击的概括性.
- 在多个基准和网络框架上的实验验证实了拟议方法的有效性.
- 证明缓解一致类混会导致更强大的DNN.
结论:
- 一致的类混 (3C) 预期提供了一个统一的视角来防御各种对抗性攻击.
- M3C方法为开发更具普遍性和强大的对抗防御策略提供了有希望的方向.
- 这项研究为克服深度学习中训练不可知攻击的挑战开辟了新的途径.
相关概念视频
Mismatch Repair
6.3K
Organisms are capable of detecting and fixing nucleotide mismatches that occur during DNA replication. This sophisticated process requires identifying the new strand and replacing the erroneous bases with correct nucleotides. Mismatch repair is coordinated by many proteins in both prokaryotes and eukaryotes.
The Mutator Protein Family Plays a Key Role in DNA Mismatch Repair
The human genome has more than 3 billion base pairs of DNA per cell. Prior to cell division, that vast amount of genetic...
The Mutator Protein Family Plays a Key Role in DNA Mismatch Repair
The human genome has more than 3 billion base pairs of DNA per cell. Prior to cell division, that vast amount of genetic...
6.3K
Strategies for Assessing and Addressing Confounding
360
Confounding is a critical issue in epidemiological studies, often leading to misleading conclusions about associations between exposures and outcomes. It occurs when the relationship between the exposure and the outcome is mixed with the effects of other factors that influence the outcome. Given that, addressing confounding is of high importance for drawing accurate inferences in research.
Confounding can be addressed at both the design phase of a study and through analytical methods after data...
Confounding can be addressed at both the design phase of a study and through analytical methods after data...
360
Conservative Site-specific Recombination and Phase Variation
6.7K
Because the DNA segments are cut and reorganized in a direction-specific manner, site-specific recombination has emerged as an efficient genetic engineering technique. Flippase and Cyclization recombinases or Flp and Cre, respectively, are two members of the tyrosine recombinase family derived from bacteriophages, that are used to mediate site-specific DNA insertions, deletions, and targeted expression of proteins in mammalian cell lines.
The recognition sites for Cre recombinase called LoxP...
The recognition sites for Cre recombinase called LoxP...
6.7K
Cognitive Dissonance
36.8K
Social psychologists have documented that feeling good about ourselves and maintaining positive self-esteem is a powerful motivator of human behavior (Tavris & Aronson, 2008). In the United States, members of the predominant culture typically think very highly of themselves and view themselves as good people who are above average on many desirable traits (Ehrlinger, Gilovich, & Ross, 2005). Often, our behavior, attitudes, and beliefs are affected when we experience a threat to our...
36.8K
Masking and Demasking Agents
3.4K
EDTA titrations may necessitate masking and demasking agents to temporarily protect a particular metal ion in a mixture from the EDTA reaction. These agents facilitate the sequential analysis of the metal ions by forming stable complexes with some—but not all—metal ions during certain steps.
There are many masking agents, such as cyanide, fluoride, triethanolamine, thiourea, and 2,3-bis(sulfanyl)propan-1-ol (formerly 2,3-dimercapto-1-propanol), with the masking agent chosen based on...
There are many masking agents, such as cyanide, fluoride, triethanolamine, thiourea, and 2,3-bis(sulfanyl)propan-1-ol (formerly 2,3-dimercapto-1-propanol), with the masking agent chosen based on...
3.4K
Types of Errors: Detection and Minimization
10.0K
Error is the deviation of the obtained result from the true, expected value or the estimated central value. Errors are expressed in absolute or relative terms.
Absolute error in a measurement is the numerical difference from the true or central value. Relative error is the ratio between absolute error and the true or central value, expressed as a percentage.
Errors can be classified by source, magnitude, and sign. There are three types of errors: systematic, random, and gross.
Systematic or...
Absolute error in a measurement is the numerical difference from the true or central value. Relative error is the ratio between absolute error and the true or central value, expressed as a percentage.
Errors can be classified by source, magnitude, and sign. There are three types of errors: systematic, random, and gross.
Systematic or...
10.0K


