If a self driving car sees a bicycle on the freeway, and it is not able to infer that the bicycle may be painted on a truck, it will be restricted to a recognition policy that is constrained by the probability of bicycles on freeways vs reliability of its object recognition.
-
-
Replying to @Plinz
What are a human's constraints in this scenario by contrast?
1 reply 0 retweets 0 likes -
Replying to @MatMcGann
Humans know that they always live in a possible world, which does not have to be a probable one.
2 replies 1 retweet 5 likes -
-
Replying to @MatMcGann @Plinz
Humans recognize a bicycle primarily from its parts (bottom up), not from its environment (top down). The latter also helps but not the primary mechanism.
1 reply 0 retweets 0 likes -
Replying to @ulusdd @MatMcGann
Most perceptual processing happens top down, it seems.
1 reply 0 retweets 0 likes -
Replying to @Plinz @MatMcGann
I am no expert in cognitive sciences but I know, from CS, top-down is usually slower, hard-to-parallelize, etc. That's bad in terms of evolution. I think it's sequential computation (of low-level pattern detectors) mistaken by top-down computation.
1 reply 0 retweets 0 likes
Perception is mostly predictive, i.e. based on complex hierarchical hypotheses about what we will perceive.
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.