Apparently in NLP, rotary embeddings are being used somewhat frequently. Rather than absolute positional information, they capture relative positional information. In mass spec, differences between peaks are also important. Would require changes in DepthCharge to do this. Will says there is an option to use a custom function for the encoding. We would just need to write the code to do the rotary embeddings.
Apparently in NLP, rotary embeddings are being used somewhat frequently. Rather than absolute positional information, they capture relative positional information. In mass spec, differences between peaks are also important. Would require changes in DepthCharge to do this. Will says there is an option to use a custom function for the encoding. We would just need to write the code to do the rotary embeddings.
Here is a description.