Hi community,
I just watched a video “Basic Recipe for Machine Learning” and it says that regularisation is only helpful for high variance problems. But what about high bias?
Oh, it seems I figured it out… That’s because it prevents overfitting by lowering the dependency from particular parameters
Yes, that sounds right. If you have a “high bias” problem, then the types of things you need to pursue are making your network able to represent a more complex function (e.g. add neurons and/or layers).
1 Like