450 points by optimization_expert 2 years ago flag hide 16 comments
optimizationguru 2 years ago next
Fascinating approach! I'm really curious to see how well this works in real-world applications.
implementationqueen 2 years ago next
@OptimizationGuru I agree! I'm excited to see if this stacks up against some of the best existing algorithms. Will you test against COIN-OR, Gurobi, or any other well known solver?
codewhisperer 2 years ago prev next
This is amazing! Does it scale to higher dimensions or specific classes of problems? I'd love to try it in my ML pipelines.
mathgenius 2 years ago next
@CodeWhisperer Yes, it does! Linear, quadratic, and even integer programming can benefit from this technique. Check out our paper for more details.
juniordeveloper 2 years ago prev next
I have terrible experiences with optimization problems in my work. I must admit that the theory has bored me to death before. Is there a blog post or video demo that can help someone with a limited math background like me?
mathgenius 2 years ago next
@JuniorDeveloper Yes, we have illustrated the approach in a Colab notebook and published a short video tutorial. It's linked from our project page along with a blog post. I hope it can help you grasp the simplicity of the algorithm.
computationalwizard 2 years ago prev next
I've heard of this approach before in some quantum computing circles. Glad to see a classical solution! Are there any parallelism-related speedup benefits we can expect?
optimizationguru 2 years ago next
@ComputationalWizard You're right about quantum computing. We've seen a good amount of parallelism benefits, getting a theoretical speedup ratio up to k times with k CPUs. More experiments need to be conducted, though.
optimizationguru 2 years ago prev next
Our initial results show promising signs against some of the known solvers, but we'll run more comparisons in the near future. Stay tuned for benchmarks! :)
quantum_programmer 2 years ago prev next
This definitely holds a lot of potential for quantum annealing. Are there any plans to test this new algorithm for quantum computers?
optimizationguru 2 years ago next
@quantum_programmer We see this approach as universal, and it's absolutely possible to port it to quantum implementations. We will explore quantum annealing as an extension to this research.
hardwarelover 2 years ago prev next
While we're exploring new ways of solving these optimization problems, will there be efficient implementations of the algorithm for both CPUs and GPUs in the future? Both markets could benefit!
optimizationguru 2 years ago next
@HardwareLover Absolutely! We're planning on open-sourcing our implementations, and our initial support will be for a broad range of platforms, including both CPUs and GPUs.
datascientist 2 years ago prev next
I've been in the field for quite a while, and I am not easily impressed by new research. However, your team seems to deserve a pat on the back. Good job!
opensourceenthusiast 2 years ago prev next
(cc: OptimizationGuru) Will you create a public Github repo for the project once it's ready to give back to the community?
optimizationguru 2 years ago next
@OpenSourceEnthusiast Yes, we love open-source as much as our fellow community members. We will create a public repository and will ensure it's clean, organized, and fun to explore!