IMPACTX: improving model performance by appropriately constraining the training with teacher explanations

Abstract

The eXplainable Artificial Intelligence (XAI) research predominantly concentrates on providing explanations about AI model decisions, especially Deep Learning (DL) models. However, there is a growing interest in using XAI techniques to automatically improve the performance of the AI systems themselves. This paper proposes IMPACTX, a novel approach that leverages XAI as a fully automated attention mechanism, without requiring external knowledge or human feedback. Experimental results show that IMPACTX improves predictive performance compared to standalone baseline models by integrating XAI-based supervision provided by teacher explanations into the training process. Furthermore, IMPACTX directly provides new feature attribution maps for the model’s decisions, without relying on external XAI methods during the inference process. Our proposal is evaluated using three widely recognized DL models (EfficientNet-B2, MobileNet, and LeNet-5) and on six publicly available benchmark datasets, comprising three image datasets (CIFAR-10, CIFAR-100, and STL-10) and three tabular datasets (Covertype, CDC Diabetes Health Indicators, and Adult). The results show that IMPACTX consistently improves the performance of all the inspected DL models across all evaluated datasets, and it directly provides appropriate explanations for its responses.