search

zhu-xlab / DeCUR

Decoupling common and unique representations for multimodal self-supervised learning
Apache License 2.0
40 stars 2 forks source link

"dim_common" parameter #6

Open weiclaim opened 9 hours ago

weiclaim commented 9 hours ago

Thank you for your wonderful work, what are the rules for setting the "dim_common" parameter?

wangyi111 commented 7 hours ago

thanks for your interest! This was a hyperparameter to tune, and we did a grid search to find the best value. In general it's taking most of the dims, so starting from e.g. 90% could be a good choice.