Bayesian Experimental Design for Symbolic Discovery
This work addresses symbolic discovery, a domain-specific problem in machine learning for model inference, but appears incremental as it applies existing Bayesian and optimization techniques without claiming major breakthroughs.
The study tackled the problem of inferring predictive models with general functional forms from observational data by applying Bayesian optimal experimental design, using constrained first-order methods and Hamiltonian Monte Carlo for optimization and sampling, with predictive distribution computed via numerical integration or fast transform methods.
This study concerns the formulation and application of Bayesian optimal experimental design to symbolic discovery, which is the inference from observational data of predictive models taking general functional forms. We apply constrained first-order methods to optimize an appropriate selection criterion, using Hamiltonian Monte Carlo to sample from the prior. A step for computing the predictive distribution, involving convolution, is computed via either numerical integration, or via fast transform methods.