Any computer vision system that works, no matter how artificial, is ipso facto a better model of human vision than any we currently have.
-
This style of argument was crushed in the 1980s because no causal or explanatory insight could be drawn from “expert system” (I.e., high performance) programs. Remember the Declarative vs Procedural controversy? end/
My point is just that by far the most important (and hardest) thing a model of vision has to do is see. If we can get there, we have a better starting point than with 1000 models that don’t.
-
-
But what does it mean to "see"? We have methods that work on benchmarks but their brittleness demonstrates that they don't see the way we do.
@aleks_madry et al. have some interesting hypotheses here:https://arxiv.org/abs/1905.02175 -
Those methods don’t really see. We don’t have any that do yet, or a good operational definition of seeing. But that should be our focus, as opposed to playing a losing game of 20 questions with Nature (cf. Newell).
- Näytä vastaukset
Uusi keskustelu -
Lataaminen näyttää kestävän hetken.
Twitter saattaa olla ruuhkautunut tai ongelma on muuten hetkellinen. Yritä uudelleen tai käy Twitterin tilasivulla saadaksesi lisätietoja.