Active Learning Reliability (Number of clusters K)

Good Morning All,

Thank you for the useful software. For Active learning Reliability, how do one determine the number of clusters ( I mean ‘K’) for enrichment and metamodel development / reliability analysis. Is there a rule of thumb?


Dear @aokoro

There is no thumb rule. The optimal convergence would be to add the points one by one. However if you can distribute computations on a cluster, i.e. run K computations in parallel, then in terms of physical time, you can add K samples at a time in a single step of the active learning. If you have a model in Matlab and a multicore processor, you can already use this functionality with a parfor loop (e.g. K=4).
Best regards