brain-research / realistic-ssl-evaluation

Open source release of the evaluation benchmark suite described in "Realistic Evaluation of Deep Semi-Supervised Learning Algorithms"
Apache License 2.0
458 stars 98 forks source link

L1/L2 regularization not found in the code #25

Closed aretor closed 5 years ago

aretor commented 5 years ago

Hi guys, In the paper it is stated that you use L1 and L2 regularizations, however I haven't found them in the code. Am I missing something? Thanks in advance

Jongchan commented 5 years ago

I couldn't find any regularization codes, too. However, I was able to achieve the reported number in the paper (CIFAR100 in WRN28-2 with 4,000 labeled samples). The paper says 20.26 +- 0.38, and I've got 19.54%. And, I've got VAT+EM 12.75% (13.13+-0.39 on paper).

avital commented 5 years ago

Apologies for the late reply. That's right, there's no L1 regularization. We should fix the paper.

avital commented 5 years ago

I submitted a new version to arXiv that no longer talks about L1 and L2 regularization.