documentation/meme_search_engine/pca.myco
2024-09-02 17:05:59 +00:00

1 line
420 B
Plaintext

osmarks.net research teams performed PCA on the matrix of meme embeddings, generating many axes of varying interpretability. The results can be found [[https://datasets.osmarks.net/components.html|here]]. Some of the more obvious axes were incorporated into the frontend as sliders. [[Sparse autoencoders]] are being investigated as an alternative if relevant datasets can [[Meme Search Engine/Reddit Dump|be acquired]].