TMLR Journal 2026 Journal Article
An Efficient Subset Selection Strategy Using Text-Guided Data Attribution to Mitigate Simplicity Bias
- Kumar Shubham
- Pranav Sastry
- Prathosh AP
The effectiveness of deep learning models heavily relies on the quality and diversity of their training data. However, datasets collected from different sources often introduce simplicity biases, where a models rely on easily learnable but non-predictive (spurious) features for its predictions. While existing debiasing techniques focus on model robustness, they leave the data untouched. However, as data becomes increasingly valuable, identifying and mitigating bias directly at the data level has become increasingly important. Recently, data attribution has emerged as a promising tool for uncovering issues in training data, yet its vulnerability to simplicity bias has received limited attention. In this work, we propose a novel data deletion framework that combines Neural Tangent Kernel (NTK)-based data attribution with textual descriptions of bias to identify and remove training samples that do not significantly affect model performance. We first demonstrate that NTK-based data attribution methods can themselves be influenced by spurious features. Subsequently, to mitigate this, we use available metadata or, when unavailable, a vision-language model to annotate a small validation set and extract a textual description of the bias. Based on this description and the attribution score, we identify the subset of training data that are semantically aligned with the spurious feature and affect the generalization of the model. Removing these samples from the training dataset and training model on the new subset improves the average and worst-group accuracy of the model, outperforming existing attribution-based baselines.