Timezone: »

 
Robust Learning with Progressive Data Expansion Against Spurious Correlation
Yihe Deng · Yu Yang · Baharan Mirzasoleiman · Quanquan Gu

@
in The Second Workshop on Spurious Correlations, Invariance and Stability »
Event URL: https://openreview.net/forum?id=CoE3nJX5C4 »

While deep learning models have shown remarkable performance in various tasks, they are susceptible to learning non-generalizable spurious features rather than the core features that are genuinely correlated to the true label. In this paper, beyond existing analyses of linear models, we theoretically examine the learning process of a two-layer nonlinear convolutional neural network in the presence of spurious features. In light of our theory, we propose a new training algorithm called PDE that efficiently enhances the model's robustness for a better worst-group performance. PDE begins with a group-balanced subset of training data and progressively expands it to facilitate the learning of the core features. Experiments on synthetic and real-world benchmark datasets confirm the superior performance of our method on models such as ResNets and Transformers.

Author Information

Yihe Deng (University of California, Los Angeles)
Yu Yang (University of California, Los Angeles)
Baharan Mirzasoleiman (UCLA)
Quanquan Gu (University of California, Los Angeles)

More from the Same Authors