@Ylecun perpetuates false dichotomy here, between nature & nurture, without presenting data. Is there any evidence primates have fewer innate priors than mountain goats? Any proof that more innate priors actually entails less adaptivity? Better priors can yield *better* learning.https://twitter.com/ylecun/status/1180171721481113602 …
-
More innate priors can't mean faster learning. We all behave reliably stupid in the middle of a town when leaping back automatically due to a 'snake-like' twig on the pavement. Zero faster learning, here. Just the contrary. Faster learning and less adaptivity! How's that?
Different priors have different consequences in different environments -- that's what the no free lunch theorem is about. [2/2] Convolution wouldn't work in a randomly ordered world without meaningful contiguity; in our world it is much better to have it than not.
-
-
here is result from LeCun 1989, decisively showing opposite of what he alleged today:pic.twitter.com/t0wULLGuRB
-
of course that is for one specific environment, per my comment above
- 1 more reply
New conversation -
-
-
Convolutions are a result of the networked computation, thresholds, frequencies and integrations. All living creatures use similar principles. It depends on the niche if it makes "sense", as that's where they are shaped. Change in environment makes the best prior meaningless, no?
-
the no free lunch theorem shows there is no best possible prior for all possible worlds. but we live in a particular world with a lot of systematic structure, & there can be best priors for specific worlds. mountain goat's priors are better than no priors would be for its world.
- 3 more replies
New conversation -
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.