Bayesian methods are not new to the HPC industry; however, IBM’s Bayesian Optimisation Accelerator (BOA) combines machine learning with a Bayesian optimisation tool. This homegrown statistical analytics stack is unique compared to Freeware for Bayesian Analysis.
This field of mathematics which has been applied to algorithms itself, does not involve machine learning. Therefore, IBM combines these to drive ensembles of HPC simulations and models.
Unlike other search methods, BOA selects the optimum solution more frequently, ensuring better business outcomes. Clients using BOA can, therefore, not only achieve the best design inside the design space but will require fewer simulations to achieve the optimal result. These benefits are particularly needed as more detailed and complex analysis is necessary with markets demanding faster answers. This state-of-the-art optimisation tool can alleviate issues arising from increasing demands without large expansions of budgets. Therefore, ensuring better business outcomes and increased productivity of experimental infrastructure without adding specialised data scientists.
IBM BOA can be used in many different areas:
Texas Advanced Computing Centre used BOA appliance in simulations. Using their standard technique for maximising oil extraction from a reservoir, it would take on the order of 200 evaluations of the model to get a good result. However, by injecting BOA into the flow of simulations, the same results were achieved using only 73 evaluations. This is a 63.5% reduction in the number of evaluations performed.
IBM’s own Power10 design team used BOA in its electronic design automation workflow. Using raw EDA software to check the signal integrity of the design took over 5,600 simulations. When IBM added BOA to the stack, only 140 simulations were required to reach the same level of accuracy. This is a 97.5% reduction in the computing needed.
A Petroleum Company using BOA found an unusual configuration that the company’s scientists had not considered. BOA was then able to find the right mix with four orders of magnitude more certainty than previous ensemble simulations. This result was found using one-third as many simulations.
As demonstrated by these case studies, innovators using IBM’s BOA can expect:
- Superior results – Locates the optimum solution while avoiding bias
- Fast innovation – BOA finds solutions faster with ease
- Fewer resources required – Methods can be applied without specialised data science skills, making existing infrastructure more efficient while keeping costs down
