Reinforcing Pre-trained Models Using Counterfactual Images
This addresses vulnerabilities in classification models for applications where reliability is critical, though it is incremental as it builds on existing pre-trained models and counterfactual methods.
The paper tackles the problem of deep learning classification models learning spurious correlations from training data by proposing a framework that uses language-guided generated counterfactual images to identify and reinforce model weaknesses, resulting in effective strengthening of models with a small set of such images.
This paper proposes a novel framework to reinforce classification models using language-guided generated counterfactual images. Deep learning classification models are often trained using datasets that mirror real-world scenarios. In this training process, because learning is based solely on correlations with labels, there is a risk that models may learn spurious relationships, such as an overreliance on features not central to the subject, like background elements in images. However, due to the black-box nature of the decision-making process in deep learning models, identifying and addressing these vulnerabilities has been particularly challenging. We introduce a novel framework for reinforcing the classification models, which consists of a two-stage process. First, we identify model weaknesses by testing the model using the counterfactual image dataset, which is generated by perturbed image captions. Subsequently, we employ the counterfactual images as an augmented dataset to fine-tune and reinforce the classification model. Through extensive experiments on several classification models across various datasets, we revealed that fine-tuning with a small set of counterfactual images effectively strengthens the model.