Hacker News Reader

Comment by GistNoesis | Hacker News Reader

Let's say you discover a new technique to solve a generic problem like Linear Programming faster for bigger instances (like PDLP). How much is it worth ? How do you turn it into cold hard cash ? How do you work developing it further with other people without revealing your secret sauce ?

View on HN

If you can beat the best solvers out there by a significant margin, I would say it's worth 10's of thousands a year. Speed is one aspect of it, but it could also be you are able to solve problems other solvers cannot solve. Depending on which companies you are targeting, I would not necessarily recommend going the route where the user upload their data to you and you solve it for them as many companies cannot legally share their clients' data (which is often part of the optimization). You can package the software into a linux/windows executable and sell it at a yearly subscription. It's worth noting that there are companies that will not be able to switch easily as they build all of their models around a specific solver API (say gurobipy)

LPaaS?

Optimisation fits the service model neatly: Submit your LP instance in some standard format, wait while it chugs away, then download the result (objective function value and variable assignments).

If you (or your client) are ok with your data going off premises.

True. It might be possible to obfuscate in various ways, e.g., by replacing each variable with k fresh variables and only caring about their sum. But this makes your instance bigger (thus more expensive) and is maybe undone anyway by all their clever presolve tricks.

One day homomorphic encryption will solve all of this properly, but my understanding is we're a long way from that.

Proprietary optimization software is very niche and very expensive but I really doubt a single innovation, unless really significant, could justify the cost.

Being able to solve bigger problems faster than someone else might give you a competitive advantage in some existing business but wouldn’t be the business in and of itself.

Maybe ask your customer how much does it worth to them? You don't have to reveal the secret sauce, provide an API to your server to keep the algorithm IP. Of course there might be race to the bottom eventually like any busniess with zero marginal cost

Just share it for free. Science doesn't work that way.

Reply on news.ycombinator.com