Bayesian Active Learning for Structured Output Design

TL;DR

This paper introduces a Bayesian active learning approach for inverse problems involving structured outputs, utilizing new acquisition functions to efficiently find inputs that produce desired multi-output results, demonstrated on synthetic and real materials data.

Contribution

It presents a novel active learning method with specialized acquisition functions for structured output inverse problems, improving efficiency in multi-output prediction tasks.

Findings

Effective in synthetic shape search problems

Successfully applied to silicon carbide crystal growth modeling

Reduces number of samples needed for desired output

Abstract

In this paper, we propose an active learning method for an inverse problem that aims to find an input that achieves a desired structured-output. The proposed method provides new acquisition functions for minimizing the error between the desired structured-output and the prediction of a Gaussian process model, by effectively incorporating the correlation between multiple outputs of the underlying multi-valued black box output functions. The effectiveness of the proposed method is verified by applying it to two synthetic shape search problem and real data. In the real data experiment, we tackle the input parameter search which achieves the desired crystal growth rate in silicon carbide (SiC) crystal growth modeling, that is a problem of materials informatics.