3 ms·
Anyone trying to use machine learning to make a "black box" mathematical framework? Feeding known measurements as "learning" (including known anomalies and disc
by jumpywizard 8y ago
Anyone trying to use machine learning to make a "black box" mathematical framework? Feeding known measurements as "learning" (including known anomalies and discrepancies)?
Using techniques such as deep learning for program synthesis. https://www.microsoft.com/en-us/research/blog/deep-learning-program-synthesis/ https://www.microsoft.com/en-us/research/blog/deep-learning-...
- pp19dd 8y agoYep, it's been done before by Cornell researchers, in 2009. You provide a bunch of data to a program and it hums for awhile and derives some equations from the observation. In the story written around the time, it took positions of a pendulum swing and derived laws of motion from it. See https://www.wired.com/2009/04/newtonai/ https://www.wired.com/2009/04/newtonai/ (and lots of other articles around the time.) You could feed a spreadsheet into and then it would iterate for awhile, converging toward most accurate equations describing the system. Or in my case, diverging from what I was sure was the solution. It was a hit and miss with most data (honestly, an eyeball and a few braincells did better with some data sets), but all the same I used it to lazily approximate position equations for some programs I was writing. The desktop program was called Eureqa which then got ported to the cloud for obvious capacity increases, and relabeled into a company/product called Nutonian. Most recently bought by "DataRobot" and it's now being sold for sales optimizations. Because science.