Re:Machine learning saves us $1.7M a year on document previews
by sk5t on Wednesday January 27, 2021 @10:42PM
(#25937867)
Setting aside the Dropbox-proprietary services, I'd be most interested to see what features/categories you picked from the file metadata, how those were prepped or rotated for training, what training/prediction algorithms you tried and how you picked a winner. Also curious to see the note that observing the "reason" was important here--isn't this a case where a total black box would be sufficient?
[ Parent
| on HN ]
There are no comments on this comment yet.