Bayesian experimental design using regularized determinantal point processes
This work addresses experimental design for researchers in statistics and machine learning, offering incremental algorithmic improvements over prior methods.
The paper tackles the problem of selecting a subset of vectors for expensive measurements in Bayesian experimental design by establishing a connection with determinantal point processes, resulting in efficient algorithms that provide (1+ε)-approximations for optimal designs under four criteria with improved efficiency and applicability.
In experimental design, we are given $n$ vectors in $d$ dimensions, and our goal is to select $k\ll n$ of them to perform expensive measurements, e.g., to obtain labels/responses, for a linear regression task. Many statistical criteria have been proposed for choosing the optimal design, with popular choices including A- and D-optimality. If prior knowledge is given, typically in the form of a $d\times d$ precision matrix $\mathbf A$, then all of the criteria can be extended to incorporate that information via a Bayesian framework. In this paper, we demonstrate a new fundamental connection between Bayesian experimental design and determinantal point processes, the latter being widely used for sampling diverse subsets of data. We use this connection to develop new efficient algorithms for finding $(1+ε)$-approximations of optimal designs under four optimality criteria: A, C, D and V. Our algorithms can achieve this when the desired subset size $k$ is $Ω(\frac{d_{\mathbf A}}ε + \frac{\log 1/ε}{ε^2})$, where $d_{\mathbf A}\leq d$ is the $\mathbf A$-effective dimension, which can often be much smaller than $d$. Our results offer direct improvements over a number of prior works, for both Bayesian and classical experimental design, in terms of algorithm efficiency, approximation quality, and range of applicable criteria.