Does @Meaningness think he can teach you a method that outperforms all possible algorithms in a classification task?https://twitter.com/IntuitMachine/status/1018814460578947072 …
-
-
Replying to @Plinz @Meaningness
My impression is he thinks a single meta-algorithm that just works out the right algorithm to use is not feasible, and instead you really need to have lots of domain-specific knowledge on what algorithm to use when.
2 replies 0 retweets 1 like -
Replying to @jprwg @Meaningness
He also seems to think that this meta knowledge acquisition can itself not be algorithmized. That in turn would mean that minds do some nebulous magic at the top level, which has self contradictory consequences if you think about it.
1 reply 0 retweets 0 likes -
Replying to @Plinz @Meaningness
It's not that it can't be algorithmised, but that doing so is glacially slow relative to a system that encodes lots of useful concepts directly. The meta-algorithm to build up this knowledge is then implemented by evolution, not individual brains.
1 reply 0 retweets 0 likes -
Replying to @jprwg @Meaningness
Evolution is basically a glacially slow random walk search algorithm. We can certainly do better.
2 replies 0 retweets 0 likes
Our brain areas may be the result of evolutionary priming, but also simply the functional implementations of the 50 dominant principal components of our cognitive domain. A ML system can look at more training data in a week than a human in a lifetime.
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.