Also E.g. an ImageNet model will be superhuman with dog types but classify people as seatbelts, and that’s okay - working as intended.
-
-
-
Replying to @marek_rosa
@marek_rosa@GaryMarcus goes to my point. misunderstanding of ImageNet's labels. There's no person class. Seatbelt class has many ppl.1 reply 0 retweets 1 like -
Replying to @karpathy
@marek_rosa@GaryMarcus ie recognition part works perfect. But if ppl run ImageNet pre trained model on data you must be aware of its data2 replies 0 retweets 1 like -
Replying to @karpathy
@marek_rosa@GaryMarcus if a classifier looks at carrot and says orange, whether it "worked" depends on if carrot was a class in data.4 replies 0 retweets 0 likes
Replying to @karpathy
@karpathy @marek_rosa Real prob is equating category learning w knowledge of closed class of items + reasoning specific to supervised labels
5:06 PM - 1 Jan 2016
0 replies
0 retweets
2 likes
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.