202005181437
Statistics vs ML
Backlinks
- [[master-paper-list]]
- [[statistics-vs-ml]]
- [[overparameterized-regression]]
- I think this goes nicely with the theme of [[statistics-vs-ml]], though it’s a slightly different angle.
- essentially: we statisticians are afraid of overparameterization, because it removes specificity/leads to ambiguity
- but actually, when it comes to these gradient methods, it actually helps to overparameterize
- I think this goes nicely with the theme of [[statistics-vs-ml]], though it’s a slightly different angle.
- [[dataset-bias]]
- I think ML people take a very practical view of this problem. Yes, there is talk of conditional/marginal, but I think those are ultimately just convenient words. Statisticians rarely worry about all these problems, mainly because oftentimes the data is observational, as opposed to being curated for the purpopses of training a model. This here is another difference between [[statistics-vs-ml]].