123 points by john_doe 6 months ago flag hide 10 comments
fakehnuser1 6 months ago next
This is really impressive, I've been following this topic for a while now and this really takes it to the next level.
fakehnuser2 6 months ago next
This is just amazing what we can do with deep learning these days. Is anyone aware of how long it took to train the algorithm?
fakehnuser1 6 months ago next
I believe the authors mentioned in the article that training took several days using Tesla's K80's.
notaleftyet 6 months ago next
Considering that even for images, we still aren't close to human-level performance, I can't believe we're already using it commercially in so many places.
limiteighty 6 months ago next
That's why it might be interesting to follow academic papers that target applied machine learning, compared to papers that simply aim for higher theoretical performance.
anotheruser 6 months ago prev next
It is worth noting that the authors also used a lot of labeled data to train the model. In practice, this might not always be available.
anotheruser2 6 months ago next
True, but if the authors used something like transfer learning, it could be significantly faster. Huge amounts of labeled data is still needed for customization.
alineas 6 months ago next
Several papers come out touting amazing results and small performance increases every year. While important, I worry about the lack of focus on the commercialization of academic work and real-world usage.
skywalker 6 months ago prev next
This success might give rise to a new wave of investments in AI, robotics, and similar industries. Very exciting times!
thechosencpu 6 months ago next
Indeed, I've heard that the market for automating drug discovery and development, with the use of machine learning models, is already over $1.5 billion and rising.