Balanced Data, Imbalanced Spectra: Unveiling Class Disparities with Spectral Imbalance
CoRR(2024)
摘要
Classification models are expected to perform equally well for different
classes, yet in practice, there are often large gaps in their performance. This
issue of class bias is widely studied in cases of datasets with sample
imbalance, but is relatively overlooked in balanced datasets. In this work, we
introduce the concept of spectral imbalance in features as a potential source
for class disparities and study the connections between spectral imbalance and
class bias in both theory and practice. To build the connection between
spectral imbalance and class gap, we develop a theoretical framework for
studying class disparities and derive exact expressions for the per-class error
in a high-dimensional mixture model setting. We then study this phenomenon in
11 different state-of-the-art pretrained encoders and show how our proposed
framework can be used to compare the quality of encoders, as well as evaluate
and combine data augmentation strategies to mitigate the issue. Our work sheds
light on the class-dependent effects of learning, and provides new insights
into how state-of-the-art pretrained features may have unknown biases that can
be diagnosed through their spectra.
更多查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要