4 ms·
> See Hebbian learning The one that is not used, because it's inherently unstable? Learning using locally accessible information is an interesting approach, b
by red75prime 2y ago
> See Hebbian learning
The one that is not used, because it's inherently unstable?
Learning using locally accessible information is an interesting approach, but it needs to be more complex than "fire together, wire together". And then you might have propagation of information that allows to approximate gradients locally.
- erikerikson 2y agoIs that what they're teaching now? Originally it was not used because it was believed it couldn't learn XOR (it can [just not as perceptrons were defined]). Is there anyone in particular whose work focuses on this that you know of?
- ckcheng 2y agoOja's rule dates back to 1982? It’s Hebbian and solves all stability problems. https://en.wikipedia.org/wiki/Oja's_rule https://en.wikipedia.org/wiki/Oja's_rule