If ppl are really publishing results using just evaluation on the training data then they should stop. In the (mostly genetics) papers I read they do at least some testing on other data, perhaps smaller. Best practices, especially if you have lots of tuning
-
-
-
parameters or forking paths in your implementation decisions (as is usually the case in ai research) you need reserve evaluation data that you don't touch until you are ready for publication results.
- 1 more reply
New conversation -
-
-
oh no
-
on the other side of things, labeled data is often now partially labeled by the machine, with corrections done by a human, potentially biasing gold standards towards what ML is good at anyway
- 1 more reply
New conversation -
-
-
Fake science
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I liked how 50 yrs of data science (2017) broke out traditional academic science generative/causal methods vs (CompSci) predictive statistics and pointed at the Kaggle-like Common Task Framework as root of the ML ratcheting progress. Related?https://www.tandfonline.com/doi/full/10.1080/10618600.2017.1384734 …
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Journalists found that on their own long time ago :) Not the first winter since 50 years ago this happens !
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.