Abstract
We present a probabilistic modeling framework and adaptive sampling algorithmwherein unsupervised generative models are combined with black box predictivemodels to tackle the problem of input design. In input design, one is given oneor more stochastic "oracle" predictive functions, each of which maps from theinput design space (e.g. DNA sequences or images) to a distribution over aproperty of interest (e.g. protein fluorescence or image content). Given suchstochastic oracles, the problem is to find an input that is expected tomaximize one or more properties, or to achieve a specified value of one or moreproperties, or any combination thereof. We demonstrate experimentally that ourapproach substantially outperforms other recently presented methods fortackling a specific version of this problem, namely, maximization when theoracle is assumed to be deterministic and unbiased. We also demonstrate thatour method can tackle more general versions of the problem.