Yann LeCun
@ylecun
@int_16h @torchedvision @MuzafferKal_ @mengyer @yingwww_ @choang333 @KuangYilun Regularizers are just a way to specify an inductive bias.
The architecture of your model, and the optimization procedure you use already have inductive biases: they like to produce certain functions more than others.
A regularizer merely imposes another bias.
Learning theory