Thanks for sharing the code of your excellent paper!
I notice that, for more than 2 feature embeddings, current example shows we assume they have equal weights. Have you considered the case where different feature embeddings has different "importance"? For example, there are two data modalities. One is by nature more informative than another. Do you consider any algorithmic way to automatically adjust the weights?
Thanks for sharing the code of your excellent paper!
I notice that, for more than 2 feature embeddings, current example shows we assume they have equal weights. Have you considered the case where different feature embeddings has different "importance"? For example, there are two data modalities. One is by nature more informative than another. Do you consider any algorithmic way to automatically adjust the weights?
Best, Bo