A Step Towards Exposing Bias in Trained Deep Convolutional Neural Network Models
This work addresses the issue of understanding and mitigating bias in deep CNN models, particularly for applications like medical imaging, by providing improved visualization tools to inform training set decisions, though it appears incremental as it combines existing techniques.
The paper tackles the problem of exposing bias in trained deep convolutional neural network models by introducing Smooth Grad-CAM++, a technique that combines SMOOTHGRAD and Grad-CAM++ to visualize layers, feature maps, or neurons, and found it produces more visually sharp maps with a larger number of salient pixels highlighted in input images compared to other methods.
We present Smooth Grad-CAM++, a technique which combines two recent techniques: SMOOTHGRAD and Grad-CAM++. Smooth Grad-CAM++ has the capability of either visualizing a layer, subset of feature maps, or subset of neurons within a feature map at each instance. We experimented with few images, and we discovered that Smooth Grad-CAM++ produced more visually sharp maps with larger number of salient pixels highlighted in the given input images when compared with other methods. Smooth Grad-CAM++ will give insight into what our deep CNN models (including models trained on medical scan or imagery) learn. Hence informing decisions on creating a representative training set.