5 ms·
DeepMind just posted a mind blowing paper – achievement of transfer learning
- ivrrimum 10y agoNice..
- skj 10y agoI wish we could keep upworthy-style titles out of academic news... Or just out of news entirely. I do my part by not reading articles with this presentation. Unless it's about celebrity twins. I'm all over that shit.
- asaddhamani 10y agoYes, please let's not have headlines like these. It might as well be "DeepMind just posted a mind blowing paper - You won't believe what happened next!" next time, so let's not go there.
- andyjohnson0 10y ago"PathNet is a network of neural networks" Can anyone comment on the reasons for this approach, compared to a single large network?
- skywhopper 10y agoOptimistic answer: each sub-network can specialize in a certain skill or element of knowledge, and then as a group, the networks can contribute to a larger decision. Such specialization allows skills to be highly developed but used only when they are applicable, instead of trying to force-fit a certain skill onto an inappropriate problem domain. The human brain is believed to work in a similar way with multiple regions processing input in parallel and "voting" for the appropriate response, with other parts of the brain coalescing and merging their output into the final response. Cynical answer: Neural net machine learning has reached its peak, and there's far more computing power than the researchers know what to do with. Add additional layers of processing and if the numbers change, that's a paper.
- PunchTornado 10y agothat article and title is the epytome of BS.
- lorenzhs 10y agoPrevious discussion of the paper: https://news.ycombinator.com/item?id=13675891 https://news.ycombinator.com/item?id=13675891 (6 comments) ...and of this very blog post: https://news.ycombinator.com/item?id=13674181 https://news.ycombinator.com/item?id=13674181 (also 6 comments)
- pizza 10y agothe paper itself https://arxiv.org/pdf/1701.08734.pdf https://arxiv.org/pdf/1701.08734.pdf
- dangom 10y ago"We can imagine that in the future, we will have giant AIs trained on thousands of tasks and able to generalize. In short, General Artificial Intelligence." Don't know if I believe this just yet.
- Grangar 10y agoThey're gonna be yuge! The greatest AIs we've ever seen!
- mrfusion 10y agoI wasn't able to follow this. Does anyone have a good explanation? Is it really AGI?
- lorenzhs 10y agoOf course not. The article is just oversensationalised. The paper doesn't make the same broad claims as the article.
- torrent-of-ions 10y agoGrammatical error in the first line. Good start.