Deep generative model for non-convex constraint handling

摘要: In this study, we consider black-box minimization problems with non-convex constraints, where the constraints are significantly cheaper to evaluate than objective. Non-convex generally make it difficult solve using evolutionary approaches. paper, revisit a conventional technique called decoder constraint handling, which transforms feasible domain into an easy-to-control convex set. This approach is promising because constrained problem almost unconstrained one. However, its application has been considerably limited, designing or training such nonlinear requires knowledge manually prepared data. To fully automate design, use deep generative models. We propose novel scheme train model without For purpose, first solution samplers, neural networks, functions. Subsequently, another data generated from trained samplers as The proposed framework applied tasks inspired by topology optimization problems. empirical study demonstrates that can locate better solutions fewer objective function evaluations existing approach.