3 ms·
While that’s the poster dream, I wonder whether it actually works in practice, rather than being a liability, as a sibling comment points out. IME, imposing do
by ssivark 6y ago
While that’s the poster dream, I wonder whether it actually works in practice, rather than being a liability, as a sibling comment points out.
IME, imposing domain knowledge (usually directly into the model, or sometimes also in the training data distribution) is crucial to getting things working in an efficient and robust manner.
- smallnamespace 6y agoDeep NNs are already competitive with specialized pipelines for translation and speech. The primary downside is that they require more data to train.
- ssivark 6y agoSure, but that’s such a tiny corner among generic problems. To generalize from there seems unwarranted.
- peterwoerner 6y agomore importantly, thanks to the internet there is a huge amount of available data that is reasonably available. Many of the science problems I have dealt with have something like 11 data "points". And experiments may take a month to perform.