Spurious correlations in the data, where multiple cues are predictive of the
target labels, often lead to a phenomenon known as simplicity bias, where a
model relies on erroneous, easy-to-learn cues while ignoring reliable ones. In
this work, we propose an ensemble diversification framework exploiting
Diffusion Probabilistic Models (DPMs) for shortcut bias mitigation. We show
that at particular training intervals, DPMs can generate images with novel
feature combinations, even when trained on images displaying correlated input
features. We leverage this crucial property to generate synthetic
counterfactuals to increase model diversity via ensemble disagreement. We show
that DPM-guided diversification is sufficient to remove dependence on primary
shortcut cues, without a need for additional supervised signals. We further
empirically quantify its efficacy on several diversification objectives, and
finally show improved generalization and diversification performance on par
with prior work that relies on auxiliary data collection.

在这项工作中，我们提出了一种利用扩散概率模型（DPMs）来减轻简洁性偏见的集成多样化框架。我们展示了在特定训练间隔中，DPMs 能够生成具有新颖特征组合的图像，即使在训练时显示相关输入特征的图像。我们利用这一关键属性通过集成差异来生成合成的反事实情景，从而增加模型的多样性。我们展示了 DPM 引导的多样化足以去除对主要简洁性提示的依赖，而无需额外的监督信号。我们进一步通过几个多样化目标进行了实证量化，并最终展示了与依赖辅助数据收集的先前工作相当的改进的泛化和多样化性能。