The SUMO toolbox: a tool for automatic regression modeling and active learning

Many complex, real world phenomena are difficult to study directly using controlled experiments. Instead, the use of computer simulations has become commonplace as a feasible alternative. Due to the computational cost of these high fidelity simulations, surrogate models are often employed as a drop-in replacement for the original simulator, in order to reduce evaluation times. In this context, neural networks, kernel methods, and other modeling techniques have become indispensable. Surrogate models have proven to be very useful for tasks such as optimization, design space exploration, visualization, prototyping and sensitivity analysis. We present a fully automated machine learning tool for generating accurate surrogate models, using active learning techniques to minimize the number of simulations and to maximize efficiency.