通过减轻硬样本干扰来提高对抗性训练.
IEEE transactions on neural networks and learning systems
|January 6, 2026
概括
缓解硬样本干扰 (MHSI) 通过提高模型准确性和稳定性来增强对抗训练 (AT). 这种新的方法减少了难以训练的样本的负面影响,从而在具有挑战性的数据集上实现了卓越的性能.
科学领域:
- 机器学习 机器学习
- 计算机视觉 计算机视觉
- 深度学习 (Deep Learning) 是一种深度学习.
背景情况:
- 反对训练 (AT) 提高了模型的稳定性,但往往牺牲了准确性.
- 共同优化准确性和稳定性可能会导致权衡,特别是在接近决策边界的硬样本.
研究的目的:
- 提出一种新的方法,即缓解硬样本干扰 (MHSI),以减轻对抗训练中的准确性-稳定性牺牲.
- 为了减少在AT期间由硬样品引起的不稳定性,而不会影响性能.
主要方法:
- 引入了加权自适应 (WA) 机制,以加强从清洁样本的学习并减少硬样本的影响.
- 设计了一种动态校准 (DC) 策略,分析梯度规范和赫森矩阵,以减轻硬样本对强度的损害.
- 从采样干预的角度实施了MHSI,以解决准确性-稳定性权衡问题.
主要成果:
- MHSI显著提高了准确性和稳定性,在玻璃盒攻击下超过了最先进的方法.
- 在CIFAR-10,CIFAR-100,Tiny ImageNet和SVHN数据集上证明了有效性.
- 在$l_{\infty }$攻击下,在AT基线上实现了6.22%的稳定性增长.
结论:
- 通过提高准确性和稳定性,MHSI有效地提高了对手训练.
- 拟议的WA和DC机制成功地减轻了硬样品干扰.
- MHSI为开发更强大,更准确的深度学习模型提供了一个有希望的方向.
相关概念视频
Improving Translational Accuracy
14.1K
Base complementarity between the three base pairs of mRNA codon and the tRNA anticodon is not a failsafe mechanism. Inaccuracies can range from a single mismatch to no correct base pairing at all. The free energy difference between the correct and nearly correct base pairs can be as small as 3 kcal/ mol. With complementarity being the only proofreading step, the estimated error frequency would be one wrong amino acid in every 100 amino acids incorporated. However, error frequencies observed in...
14.1K
Improving Translational Accuracy
3.5K
3.5K
Strategies for Assessing and Addressing Confounding
356
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...
356
Atomic Absorption Spectroscopy: Interference
2.0K
Interference leads to systematic error in atomic absorption (AA) measurements by enhancing or diminishing the analytical signal or the background. These interferences can be grouped into three main categories: spectral interference, chemical interference, and physical interference.
Spectral interference occurs when signals from other elements or molecules overlap with the analyte signal, falsely elevating or masking the analyte's absorbance. This interference can be corrected using Zeeman,...
Spectral interference occurs when signals from other elements or molecules overlap with the analyte signal, falsely elevating or masking the analyte's absorbance. This interference can be corrected using Zeeman,...
2.0K
Halo Effect
395
The halo effect is a cognitive bias in which an individual's overall impression influences judgments about their specific traits. This psychological phenomenon leads people to associate positive characteristics with those they perceive as generally good and negative characteristics with those they view as bad. This effect is particularly influential in social perception, professional evaluations, and decision-making processes.The Psychological Basis of the Halo EffectThe halo effect is rooted...
395
Confirmation Biases
7.6K
The confirmation bias is the tendency to focus on information that confirms our existing beliefs and ignore information that is inconsistent with our expectations. For example, if you think that your professor is not very nice, you notice all of the instances of rude behavior exhibited by the professor while ignoring the countless pleasant interactions he is involved in on a daily basis. Have you ever fallen prey to the confirmation bias, either as the source or target of such bias?
7.6K
