When traditional education folks study MOOCs they totally mess it up every time.— Jeremy Howard (@jeremyphoward) February 3, 2019
e.g. Why on earth are people still using completion rate as a success metric? Our course, for instance, is explicitly designed to provide most of what you need in the first couple of lessons.
The end of the article is disappointing. The MOOCs weren't "domesticated" by existing higher ed; instead their learning model failed on its own. Would like to see a study that explained WHY they failed. The need is huge.— Thomas G. Dietterich (@tdietterich) February 3, 2019
Unless you spend a lot of effort on examining the exact characteristics of the problem (which is very difficult), you should not be claiming that your algorithm outperforms anything outside of the narrow set of benchmarks you tested on.— Denny Britz (@dennybritz) February 3, 2019
"Formulating data science problems is an uncertain and difficult process... Whether we consider a data science project fair often has as much to do with the formulation of the problem as any property of the resulting model." @s010n @samirpassi https://t.co/bV8SKSm9GM pic.twitter.com/iDfpKQ4YQU— Rachel Thomas (@math_rachel) February 1, 2019
This is a super cool resource: Papers With Code now includes 950+ ML tasks, 500+ evaluation tables (including SOTA results) and 8500+ papers with code. Probably the largest collection of NLP tasks I've seen including 140+ tasks and 100 datasets.https://t.co/lTAGE7LGZY pic.twitter.com/wfSyTplBR3— Sebastian Ruder (@seb_ruder) February 1, 2019
Loving @ThisIsSethsBlog's data thoughts. Here are mine:— Caitlin Hudon👩🏼💻 (@beeonaposy) February 1, 2019
1. Don’t prioritize analyses which have a zero chance of changing your biz / product.
2. Storage + analysis isn't free.
3. Your data collection mechanism (+ impact on accuracy) is just as important as the analyses you do. pic.twitter.com/WtNuEA9ReY