Fetching the paper…

Enhancing Neural Network Interpretability with Feature-Aligned Sparse Autoencoders · Around