Interpretability by design is usually a conscientious effort. Researchers will think of new architectures or adaptions to existing ones that … Continue reading Class Activation Maps (CAMs) from Scratch
How global average pooling layers lead to intrinsically interpretable neural networks