7 ms·
That is addressing the incomprehensibility of PCA and applying a transformation to the entire latent space. I've never found PCA to be meaningful for deep learn
by goldemerald 1y ago
That is addressing the incomprehensibility of PCA and applying a transformation to the entire latent space. I've never found PCA to be meaningful for deep learning. As far as I can tell, polysemous issue with neurons cannot be addressed with a single linear transformation. There is no sparse analysis (via linear probes or SAEs) and hence the unaddressed issue.