- Explainer: Regularization is supposed to be a compromise, between a function that incentives fitting your date, and a regularizer that wants a simple model. But the Lasso will famously toss out the goodness-of-fit concerns and give you a fully sparse/simple model. How?? Why?? #BGBlog #statsky #mlsky
Feb 2, 2026 17:20
- At Google, I'd lead intro-to-ML seminars, and I always felt guilty that we had to handwave "L1 regularization means weights that are exactly zero," with no further explanation/guidance. So I wrote a full rundown for readers who know precalc and show it really is useful to know the *why* of sparsity.
