Fast Bayesian optimization methods match SOTA at 100x–1000x speed
Original title: Fast Bayesian Optimization Through Linear Models
The methods target continuous and combinatorial inputs from about 10 to more than 6000 dimensions by using linear structure already in BO.
A paper argues Bayesian optimization is used mainly when evaluations are expensive because the optimizer itself is slow. It frames BO against faster methods such as CMA-ES and says parts of the speed-performance frontier are empty. The authors propose fast BO methods that exploit linear structure for continuous and combinatorial inputs, claiming performance similar to state-of-the-art BO at more than 100x to 1000x the speed.
Unconfirmed: Speed and performance claims are the authors' reported results, not independently verified here.
- Methods cover continuous and combinatorial inputs from about 10 to more than 6000 dimensions.
- Claimed speedup is more than 100 times to 1000 times versus existing BO.
- Performance is described as similar to state-of-the-art Bayesian optimization.
- CMA-ES is the faster baseline used in the comparison.
From the original article
1 min read at arxiv.orgBayesian optimization (BO) is typically used only for problems where the underlying function is deemed "expensive" (e. g. in time or money), with the speed of the BO algorithm being of little importance.
By Colin Doumont · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.