Bias in DeepSVDD

Created by: psorus

If you read Lukas Ruffs Paper (https://ls9-www.cs.tu-dortmund.de/publications/ICML2018.pdf ) he writes in 3.3 Proposition 2 that "Proposition 2 implies that networks with bias terms can easily learn any constant function" and thus suggests not using a bias term in the definition of a DeepSVDD paper. Also, his code (https://github.com/lukasruff/Deep-SVDD-PyTorch/blob/1901612d595e23675fb75c4ebb563dd0ffebc21e/src/networks/mnist_LeNet.py#L16 ) disables these bias terms. Is there any reason I am not aware of why pyod's version of deep svdd uses biases in its network definition?