[论文解读] Understanding CNN Fragility When Learning With Imbalanced Data

Damien Dablain, Kristen N. Jacobson+3 more|arXiv (Cornell University)|Oct 17, 2022

Imbalanced Data Classification Techniques被引用 6

一句话总结

本文通过分析潜在特征嵌入（FE）和分类嵌入（CE），研究了CNN在类别不平衡数据集上对少数类泛化能力差的原因。研究发现，模型在少数类上依赖于一组数量有限、高幅值且多样性低的特征，当测试特征的幅值与训练数据不一致时，这种依赖导致模型变得脆弱——这表明，特征多样性（而不仅仅是类别平衡）对于实现稳健泛化至关重要。

ABSTRACT

Convolutional neural networks (CNNs) have achieved impressive results on imbalanced image data, but they still have difficulty generalizing to minority classes and their decisions are difficult to interpret. These problems are related because the method by which CNNs generalize to minority classes, which requires improvement, is wrapped in a blackbox. To demystify CNN decisions on imbalanced data, we focus on their latent features. Although CNNs embed the pattern knowledge learned from a training set in model parameters, the effect of this knowledge is contained in feature and classification embeddings (FE and CE). These embeddings can be extracted from a trained model and their global, class properties (e.g., frequency, magnitude and identity) can be analyzed. We find that important information regarding the ability of a neural network to generalize to minority classes resides in the class top-K CE and FE. We show that a CNN learns a limited number of class top-K CE per category, and that their number and magnitudes vary based on whether the same class is balanced or imbalanced. This calls into question whether a CNN has learned intrinsic class features, or merely frequently occurring ones that happen to exist in the sampled class distribution. We also hypothesize that latent class diversity is as important as the number of class examples, which has important implications for re-sampling and cost-sensitive methods. These methods generally focus on rebalancing model weights, class numbers and margins; instead of diversifying class latent features through augmentation. We also demonstrate that a CNN has difficulty generalizing to test data if the magnitude of its top-K latent features do not match the training set. We use three popular image datasets and two cost-sensitive algorithms commonly employed in imbalanced learning for our experiments.

研究动机与目标

• 理解CNN为何在类别不平衡数据集中无法对少数类实现泛化。
• 研究潜在特征嵌入（FE）和分类嵌入（CE）如何影响模型决策。
• 检验CNN学习的是类别内在特征，还是仅从不平衡训练数据中学习到的频繁模式。
• 评估特征幅值与多样性在模型泛化性能中的作用。
• 挑战仅通过重平衡类别频率即可确保鲁棒性的假设，提出潜在特征多样性与类别平衡同等重要。

提出的方法

• 作者将训练好的CNN分解为特征提取头和分类头，以隔离并分析潜在表征。
• 他们从网络最后一层提取并分析特征嵌入（FE）和分类嵌入（CE）。
• 他们定义并计算每个类别的top-K FE和CE，重点关注身份、幅值、频率和密度。
• 他们通过比较训练集和测试集的嵌入，评估特征响应范围和身份的一致性。
• 他们使用FB范数量化top-K特征的幅值与多样性，并将其与泛化性能关联。
• 实验在三个数据集（SVHN、CelebA、CIFAR-10）上进行，采用两种代价敏感损失函数（焦点损失、LDAM），通过TP/FP分析评估泛化性能。

实验结果

研究问题

• RQ1 在类别不平衡设置下，多数类与少数类的top-K特征嵌入和分类嵌入在身份与幅值上存在何种差异？
• RQ2 模型泛化性能在多大程度上依赖于训练集与测试集之间特征幅值范围的一致性？
• RQ3 模型学习的是类别内在特征，还是仅从训练分布中学习到的高频模式？
• RQ4 特征多样性（top-K FE/CE的数量及其幅值分布范围）与少数类泛化能力之间有何相关性？
• RQ5 如果未解决潜在特征多样性问题，仅通过重采样或代价敏感方法能否确保鲁棒性？

主要发现

• CNN每个类别仅学习到有限数量的top-K分类嵌入，在少数类上其数量更少且幅值更低，与平衡设置相比差异显著。
• 少数类的预测依赖于一组数量稀少的高幅值特征，当测试特征幅值超出训练响应范围时，模型变得脆弱。
• 真正例（TPs）的FB范数显著低于假正例（FPs）（TPs为1.73–4.67，FPs为9.00–21.16），表明正确预测基于更一致、更高幅值的特征响应。
• 即使top-K特征身份有70%的重叠，当训练集与测试集的特征幅值范围不一致时，模型仍无法实现泛化。
• 使用交叉熵或代价敏感损失训练的模型，若测试集特征幅值不匹配训练集幅值，则无法实现泛化，表明模型已记忆特定响应范围。
• 本研究结论认为，潜在特征多样性——通过top-K嵌入的数量与幅值分布范围衡量——与类别平衡一样，对实现稳健泛化至关重要。