A neural network learns to recognize objects and outputs '0001' for a car, '0010' for a boat, etc. But it could learn to produce any output code for these objects. So where does the object recognition actually happen? I'm still the one that decides that '0001' = car.
-
Show this thread
-
Neural networks don't recognize objects, they lay down coordinate systems on high dimensional spaces. We decide to map labels to regions of these coordinate systems. We're still doing the recognition.
2 replies 1 retweet 10 likesShow this thread -
Replying to @bradtheilman
Well, it's us that decided that there's such a thing as "objects"
1 reply 0 retweets 6 likes -
Replying to @_julesh_
Yeah this is an important point... neuroscience seems to be moving in a direction that assumes ANNs have solved object recognition but forgets we are also trying to explain "objects" and "recognition"
1 reply 0 retweets 1 like -
Replying to @bradtheilman
On second thoughts, how do we know it's not the other way round? Maybe the scene really does contain an 0100, but humans are limited and so produce the output "car" which represents 0100
2 replies 0 retweets 3 likes
This is exactly what is happening
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.