Skip to yearly menu bar Skip to main content


Active Nearest Neighbor Regression Through Delaunay Refinement

Alexander Kravberg · Giovanni Luca Marchetti · Vladislav Polianskii · Anastasiia Varava · Florian T. Pokorny · Danica Kragic

Hall E #727

Keywords: [ PM: Monte Carlo and Sampling Methods ] [ MISC: General Machine Learning Techniques ] [ OPT: Zero-order and Black-box Optimization ] [ OPT: Sampling and Optimization ]


We introduce an algorithm for active function approximation based on nearest neighbor regression. Our Active Nearest Neighbor Regressor (ANNR) relies on the Voronoi-Delaunay framework from computational geometry to subdivide the space into cells with constant estimated function value and select novel query points in a way that takes the geometry of the function graph into account. We consider the recent state-of-the-art active function approximator called DEFER, which is based on incremental rectangular partitioning of the space, as the main baseline. The ANNR addresses a number of limitations that arise from the space subdivision strategy used in DEFER. We provide a computationally efficient implementation of our method, as well as theoretical halting guarantees. Empirical results show that ANNR outperforms the baseline for both closed-form functions and real-world examples, such as gravitational wave parameter inference and exploration of the latent space of a generative model.

Chat is not available.