Explaining Deep Convolutional Neural Networks for Image Classification by Evolving Local Interpretable Model-agnostic Explanations

TL;DR

This work addresses the explainability gap in deep CNN image classifiers by introducing E-LIME, a genetic-algorithm–driven local explainer that is model-agnostic and avoids the sampling and fixed-feature-count limitations of LIME. By encoding superpixel selections as binary vectors and optimizing them with a GA, E-LIME rapidly uncovers human-interpretable, image-adapted explanations that align with meaningful features (e.g., ears, eyes, beaks) and can boost the model’s predicted probabilities. The method is demonstrated on ImageNet using ResNet, DenseNet, MobileNet, and even tested on Vision Transformer, showing faster execution (within minutes) and improved confidence in predictions compared to LIME. These findings suggest practical benefits for trust, accountability, and debugging of deep CNNs in real-world image classification tasks.

Abstract

Deep convolutional neural networks have proven their effectiveness, and have been acknowledged as the most dominant method for image classification. However, a severe drawback of deep convolutional neural networks is poor explainability. Unfortunately, in many real-world applications, users need to understand the rationale behind the predictions of deep convolutional neural networks when determining whether they should trust the predictions or not. To resolve this issue, a novel genetic algorithm-based method is proposed for the first time to automatically evolve local explanations that can assist users to assess the rationality of the predictions. Furthermore, the proposed method is model-agnostic, i.e., it can be utilised to explain any deep convolutional neural network models. In the experiments, ResNet is used as an example model to be explained, and the ImageNet dataset is selected as the benchmark dataset. DenseNet and MobileNet are further explained to demonstrate the model-agnostic characteristic of the proposed method. The evolved local explanations on four images, randomly selected from ImageNet, are presented, which show that the evolved local explanations are straightforward to be recognised by humans. Moreover, the evolved explanations can explain the predictions of deep convolutional neural networks on all four images very well by successfully capturing meaningful interpretable features of the sample images. Further analysis based on the 30 runs of the experiments exhibits that the evolved local explanations can also improve the probabilities/confidences of the deep convolutional neural network models in making the predictions. The proposed method can obtain local explanations within one minute, which is more than ten times faster than LIME (the state-of-the-art method).

Figures (17)

• Figure 1: ResNet architecture he2016deep.
• Figure 2: Sample feature maps of ResNet. (a) shows the original image of a Persian cat. (b) and (c) are feature maps extracted from ResNet.
• Figure 3: Overall framework.
• Figure 4: Interpretable features vs uninterpretable feature maps.
• Figure 5: Encoding strategy.