Bayesian Optimization with Approximate Set Kernels
This work addresses a domain-specific problem in optimization for machine learning, offering incremental improvements in efficiency for set-based tasks.
The paper tackled the problem of Bayesian optimization over permutation-invariant set inputs by developing a method with an approximate set kernel and constrained acquisition function optimization, resulting in outperformance over other methods in numerical experiments.
We propose a practical Bayesian optimization method over sets, to minimize a black-box function that takes a set as a single input. Because set inputs are permutation-invariant, traditional Gaussian process-based Bayesian optimization strategies which assume vector inputs can fall short. To address this, we develop a Bayesian optimization method with \emph{set kernel} that is used to build surrogate functions. This kernel accumulates similarity over set elements to enforce permutation-invariance, but this comes at a greater computational cost. To reduce this burden, we propose two key components: (i) a more efficient approximate set kernel which is still positive-definite and is an unbiased estimator of the true set kernel with upper-bounded variance in terms of the number of subsamples, (ii) a constrained acquisition function optimization over sets, which uses symmetry of the feasible region that defines a set input. Finally, we present several numerical experiments which demonstrate that our method outperforms other methods.