250 points by opti_solvers 5 months ago flag hide 9 comments
diviner 5 months ago next
Fascinating! I've been looking for solutions in addressing large scale optimization problems. This approach could be very relevant and helpful in my field (data science). I'm especially interested in understanding the underlying complexities that enable the efficiency they've achieved.
paradoxlost 5 months ago next
I agree with you, diviner! It is really a refreshing take on things and the fact that it brings mathematical simplicity to the table makes it even more appealing. Hopefully, the authors can share a comprehensive article on the mathematical complexities soon!
sdldxl 5 months ago prev next
Great to see the combination of techniques like approximations and scalable parallel algorithms to solve large-scale problems. Kudos to the team, looking forward to learning the step-by-step process in-depth.
tomorrowlabs 5 months ago next
@sdldxl: As researchers, we'll make sure to cover the methodologies in a way that it is accessible to the community. We believe that’s the key to making real progress in AI research and fostering innovation everywhere!
hankchen 5 months ago prev next
Very excited about this the practical uses in machine learning. Wonderful to finally see great minds tackling these NP-hard problems.
er1c 5 months ago prev next
I get the sense this could've been applicable to my thesis work. Bummer to see it now when I've already submitted it!
ablandscape 5 months ago prev next
This is unlike any other solution I’ve seen so far. I'm very curious to find out more about its robustness and application range. And if there is some low hanging fruits (real-world use cases) we can gather, that’ll be awesome!
diviner 5 months ago next
@AB Landscape: While they do claim the approach is applicable to a wide variety of problems, I'd also be interested to see the robustness in a few real-world use cases first. It adds a sense of validation that can be helpful in gauging overall effectiveness. Looking forward to the tests they’ll conduct!
cyberwit 5 months ago prev next
Will there be an implementation on GitHub? It will make it easy to try this out for people experimenting on their own.