I’ve been noticing that a lot of techniques seem to repeat across modalities with just a different name.

For example, tokenisation, masking (MLM / MAE / audio masking), diffusion, next-token / next-frame prediction, etc.

Is there an existing taxonomy or mental model that groups ML techniques by their underlying pattern rather than by modality (NLP / vision / audio)?

This will help me to group the research papers and then make reference to work already done in that domain