Post tl;dr: Sometimes when you are optimizing an ML / AI pipeline you care about multiple things (speed, accuracy, etc). It is often difficult to make these tradeoffs without knowing what is possible, but it is expensive to try different things. Using Bayesian optimization to optimize multiple metrics simultaneously can help solve this problem more than an order of magnitude faster than standard techniques like randomized search. We provide examples tuning TensorFlow and MXnet CNNs with code [1] [2].
SigOpt tl;dr: Parameter optimization-as-a-service. An ensemble of Bayesian optimization techniques behind a simple REST API. Spend more time building your models while we optimally tune them exponentially more efficiently than something like a brute force grid search. Free for academic use [3].
4 comments
[ 4.5 ms ] story [ 18.1 ms ] threadMore info on the research behind SigOpt can be found here [1].
[1]: https://sigopt.com/research
SigOpt tl;dr: Parameter optimization-as-a-service. An ensemble of Bayesian optimization techniques behind a simple REST API. Spend more time building your models while we optimally tune them exponentially more efficiently than something like a brute force grid search. Free for academic use [3].
[1]: https://github.com/sigopt/sigopt-examples/tree/master/multim...
[2]: https://github.com/sigopt/sigopt-examples/tree/master/dnn-tu...
[3]: https://sigopt.com/edu