A Neurosymbolic Framework for Bias Correction in Convolutional Neural Networks
This work addresses bias correction in CNNs for image classification tasks, offering a method to enhance fairness and interpretability, though it is incremental as it builds on existing neurosymbolic approaches.
The authors tackled the problem of correcting biases in trained convolutional neural networks (CNNs) by introducing a neurosymbolic framework called NeSyBiCor, which uses semantic similarity loss to retrain CNNs based on symbolic constraints, resulting in bias correction with minimal accuracy loss and improved interpretability on the Places dataset.
Recent efforts in interpreting Convolutional Neural Networks (CNNs) focus on translating the activation of CNN filters into a stratified Answer Set Program (ASP) rule-sets. The CNN filters are known to capture high-level image concepts, thus the predicates in the rule-set are mapped to the concept that their corresponding filter represents. Hence, the rule-set exemplifies the decision-making process of the CNN w.r.t the concepts that it learns for any image classification task. These rule-sets help understand the biases in CNNs, although correcting the biases remains a challenge. We introduce a neurosymbolic framework called NeSyBiCor for bias correction in a trained CNN. Given symbolic concepts, as ASP constraints, that the CNN is biased towards, we convert the concepts to their corresponding vector representations. Then, the CNN is retrained using our novel semantic similarity loss that pushes the filters away from (or towards) learning the desired/undesired concepts. The final ASP rule-set obtained after retraining, satisfies the constraints to a high degree, thus showing the revision in the knowledge of the CNN. We demonstrate that our NeSyBiCor framework successfully corrects the biases of CNNs trained with subsets of classes from the "Places" dataset while sacrificing minimal accuracy and improving interpretability.