If you trace back from AlexNet you get to this 1979 NeoCognitron thing by Kunihiko Fukushima as the earliest candidate for ancestor of modern ML 🧐. Is this generally acknowledged? I just traced back from Wikipedia
Conversation
Replying to
It's certainly extremely well known. Earliest candidate? Certainly not, goes back decades earlier.
1
4
Replying to
Ah, so you mean early neural nets work, not ML. Work on statistical ML models of language goes back to circa 1950, for instance, and really looks remarkably modern in some ways.
2
5
Replying to
Yeah, I’ve been sloppily treating them as effectively synonymous now. Feels like the key breakthrough was multiple layers + GPU use?
Replying to
"the key breakthrough" seems like a fruitless search 😀
Backprop, Turing-completeness, the transistor, GPUs, (really) big data & other regularization techniques [etc etc etc].
1
3
Show replies
Show replies


