Daisuke Niizumi

Thanks to your CLR implementation, I forked and tailored another version for trapezoid schedule which is introduced in this paper: - [Chen Xing, Devansh Arpit, Christos Tsirigotis, Yoshua Bengio, A...

BYOL-A is incompletely implemented...

10

Dear @leo19941227 or somebody who can help, I appreciate your interest. I'm the author of the BYOL-A paper. I got to know that our BYOL-A is implemented as one of...

Hi, This is for sharing a possible issue who prefers a newer sklearn. it seems that a new member variable has been added to the MinMaxScaler. ```python model = return_loaded_model(AudioEncoder,...

Fix BYOL-A implementation

8

Hi, let me make a request to fix the BYOL-A implementation. We made the followings: 1. Fixed the usage regarding the calculation of normalization statistics. It is now required to...

リポジトリありがとうございます。実装例のおかげで手元の応用もスムーズに行きました。 Qiita記事で日本語でしたので、こちらもあえて日本語でコメントを書かせてください。 一点だけ気づいたのですが、こちらの混合比率がひょっとすると違うかもしれません。 ```Python loss_hard_target = -tf.reduce_sum(y_ * tf.log(y), reduction_indices=[1]) loss_soft_target = -tf.reduce_sum(soft_target_ * tf.log(y_soft_target), \ reduction_indices=[1]) In [16]: loss = tf.reduce_mean(\ tf.square(T) * loss_hard_target \ + tf.square(T) * loss_soft_target)...