Skip to main content
QUICK REVIEW

[论文解读] Transferred Discrepancy: Quantifying the Difference Between Representations

Yunzhen Feng, Runtian Zhai|arXiv (Cornell University)|Jul 24, 2020
Generative Adversarial Networks and Image Synthesis参考文献 28被引用 6
一句话总结

本文提出了迁移差异(Transferred Discrepancy, TD),这是一种新颖的度量方法,通过衡量两个模型在下游任务性能上的差异来量化表示差异。结果表明,TD与泛化质量具有强相关性,即提升泛化能力的数据增强方法也会增强TD的鲁棒性,而有害的数据增强则会降低TD。

ABSTRACT

Understanding what information neural networks capture is an essential problem in deep learning, and studying whether different models capture similar features is an initial step to achieve this goal. Previous works sought to define metrics over the feature matrices to measure the difference between two models. However, different metrics sometimes lead to contradictory conclusions, and there has been no consensus on which metric is suitable to use in practice. In this work, we propose a novel metric that goes beyond previous approaches. Recall that one of the most practical scenarios of using the learned representations is to apply them to downstream tasks. We argue that we should design the metric based on a similar principle. For that, we introduce the transferred discrepancy (TD), a new metric that defines the difference between two representations based on their downstream-task performance. Through an asymptotic analysis, we show how TD correlates with downstream tasks and the necessity to define metrics in such a task-dependent fashion. In particular, we also show that under specific conditions, the TD metric is closely related to previous metrics. Our experiments show that TD can provide fine-grained information for varied downstream tasks, and for the models trained from different initializations, the learned features are not the same in terms of downstream-task predictions. We find that TD may also be used to evaluate the effectiveness of different training strategies. For example, we demonstrate that the models trained with proper data augmentations that improve the generalization capture more similar features in terms of TD, while those with data augmentations that hurt the generalization will not. This suggests a training strategy that leads to more robust representation also trains models that generalize better.

研究动机与目标

  • 为解决现有表示相似性度量方法在基于特征值时产生矛盾结论的问题,即在不同情况下得出不一致的结论。
  • 提出一种基于下游任务性能的新度量方法,以反映表示在实际应用中的运作方式。
  • 在线性探测的理论框架下建立新度量方法的理论基础,并阐明其与现有度量方法(如CKA和CCA)的关系。
  • 通过实证验证,表明TD能忠实反映不同初始化和训练策略下所学习表示的、与任务相关的差异。
  • 从TD鲁棒性的视角,探究训练因素(如数据增强和模型架构)对表示相似性的影响。

提出的方法

  • 将迁移差异(TD)定义为:在对两个模型的特征进行线性头微调后,其在相同下游任务数据上预测结果的差异。
  • 使用线性探测对TD在合理变换下的不变性进行理论分析,并推导其渐近极限。
  • 证明在特定条件下,TD在数学上与现有度量方法(如最大匹配、CCA和CKA)相关。
  • 在多个下游任务(如CIFAR-10、CIFAR-5)上评估TD,使用不同数据增强、架构和上游任务训练的模型。
  • 引入并分析TD鲁棒性,即在相同下游任务上,从不同随机初始化训练的模型之间预测差异的差异。
  • 通过消融实验,考察批量大小、训练轮数、模型深度、宽度以及上游任务难度对TD敏感性和可靠性的影​​响。

实验结果

研究问题

  • RQ1如何设计一种表示相似性度量方法,使其能真实反映下游任务中的实际性能?
  • RQ2为何现有度量方法(如CKA和CCA)有时会对表示相似性得出矛盾的结论?
  • RQ3迁移差异(TD)度量在多大程度上与下游泛化性能保持一致?
  • RQ4数据增强策略在多大程度上影响通过TD度量的表示相似性?
  • RQ5架构选择(如深度和宽度)在TD度量下如何影响表示相似性?

主要发现

  • 从不同初始化训练的模型在下游任务性能上的表示差异可通过TD测量,表明其表示并非完全相同。
  • 当数据增强方法提升泛化能力时(如CutMix和MixUp),TD鲁棒性显著提高;而当增强方法损害泛化能力时,TD鲁棒性则下降。
  • 在Cifar-5上训练的ResNet110在Cifar-10上达到83%准确率,其TD值低于浅层模型,表明其表示更相似。
  • 将ResNet32的宽度加倍后,TD_hard从0.2139降至0.1580,表明宽度对表示相似性的影响强于深度。
  • 随着上游任务难度增加(C4 → C100),TD_soft和TD_hard均下降,表明更困难的上游任务会导致更相似的表示。
  • 在C100上训练的VGG19模型在Cifar-10上的TD_hard最高(0.4567),表明在某些条件下,更深的架构可能学习到更不相似的特征。

更好的研究,从现在开始

从阅读论文到最终审阅,大幅缩短您的研究时间。

无需绑定信用卡

本解读由 AI 生成,并经人工编辑审核。