Hacker News Reader

I remember ART from 25 years ago, sounded very promising even if I didn't fully understand at the time (nor now, but not revisited).

Obviously back-propagation based NN have become very popular (even got a rebrand as Deep Learning), how has ART changed in the last 25 years, is there any interesting results using ART instead of DL?

View on HN

There have been updates on the supervised versions of ART (ARTMAP) but none I know of, which have been shown to be as as effective as DL methods on very large datasets. Not sure if there have been many/any attempts to tackle large datasets. DL methods, while opaque, often result in good feature basis sets. ART given it was devised to model perception in brains, assumes features are created by lower network layers (V1, primary auditory cortex etc). I don't think there's any reason to suspect ART could not be extended to aid feature discovery, but it hasn't been done as far as I know.

Reply on news.ycombinator.com