Combining Latent Space and Structured Kernels for Bayesian Optimization over Combinatorial Spaces

We consider the problem of optimizing combinatorial spaces (e.g., sequences, trees, and graphs) using expensive black-box function evaluations.