Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhanced Bayesian Optimization via Preferential Modeling of Abstract Properties

Arun Kumar A, Alistair Shilton, Sunil Gupta, Santu Rana, Stewart Greenhill, Svetha Venkatesh

TL;DR

This work addresses the inefficiency of conventional Bayesian Optimization by incorporating human expertise on unmeasured abstract properties. It introduces BOAP, a framework that learns latent abstract-property functions via Rank Gaussian Processes from pairwise expert preferences and augments the BO surrogate with these insights, guarded by a two-arm model selection to mitigate biased guidance. Theoretical convergence discussions and empirical validation on synthetic benchmarks and real-world battery manufacturing tasks demonstrate that BOAP can achieve superior sample efficiency and robust performance when expert preferences are informative, while remaining resilient to imperfect inputs. The approach holds practical significance for accelerating expensive experimental design across engineering domains where high-level physical properties guide decision-making.

Abstract

Experimental (design) optimization is a key driver in designing and discovering new products and processes. Bayesian Optimization (BO) is an effective tool for optimizing expensive and black-box experimental design processes. While Bayesian optimization is a principled data-driven approach to experimental optimization, it learns everything from scratch and could greatly benefit from the expertise of its human (domain) experts who often reason about systems at different abstraction levels using physical properties that are not necessarily directly measured (or measurable). In this paper, we propose a human-AI collaborative Bayesian framework to incorporate expert preferences about unmeasured abstract properties into the surrogate modeling to further boost the performance of BO. We provide an efficient strategy that can also handle any incorrect/misleading expert bias in preferential judgments. We discuss the convergence behavior of our proposed framework. Our experimental results involving synthetic functions and real-world datasets show the superiority of our method against the baselines.

Enhanced Bayesian Optimization via Preferential Modeling of Abstract Properties

TL;DR

This work addresses the inefficiency of conventional Bayesian Optimization by incorporating human expertise on unmeasured abstract properties. It introduces BOAP, a framework that learns latent abstract-property functions via Rank Gaussian Processes from pairwise expert preferences and augments the BO surrogate with these insights, guarded by a two-arm model selection to mitigate biased guidance. Theoretical convergence discussions and empirical validation on synthetic benchmarks and real-world battery manufacturing tasks demonstrate that BOAP can achieve superior sample efficiency and robust performance when expert preferences are informative, while remaining resilient to imperfect inputs. The approach holds practical significance for accelerating expensive experimental design across engineering domains where high-level physical properties guide decision-making.

Abstract

Experimental (design) optimization is a key driver in designing and discovering new products and processes. Bayesian Optimization (BO) is an effective tool for optimizing expensive and black-box experimental design processes. While Bayesian optimization is a principled data-driven approach to experimental optimization, it learns everything from scratch and could greatly benefit from the expertise of its human (domain) experts who often reason about systems at different abstraction levels using physical properties that are not necessarily directly measured (or measurable). In this paper, we propose a human-AI collaborative Bayesian framework to incorporate expert preferences about unmeasured abstract properties into the surrogate modeling to further boost the performance of BO. We provide an efficient strategy that can also handle any incorrect/misleading expert bias in preferential judgments. We discuss the convergence behavior of our proposed framework. Our experimental results involving synthetic functions and real-world datasets show the superiority of our method against the baselines.
Paper Structure (38 sections, 35 equations, 6 figures, 2 tables, 3 algorithms)

This paper contains 38 sections, 35 equations, 6 figures, 2 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Background
  3. Bayesian Optimization
  4. Gaussian Process
  5. Acquisition Functions
  6. Rank GP Distributions
  7. Framework
  8. Expert Preferential Inputs on Abstract Properties
  9. Rank GPs for Abstract Properties
  10. Hyperparameter Optimization
  11. Augmented GP with Abstract Property Preferences
  12. Overcoming Inaccurate Expert Inputs
  13. Convergence Remarks
  14. Experiments
  15. Synthetic Experiments
  16. ...and 23 more sections

Figures (6)

  • Figure 1: A schematic representation of Bayesian Optimization with Abstract Properties (BOAP)
  • Figure 2: Simple regret vs iterations for the synthetic multi-dimensional benchmark functions. We plot the average regret along with its standard error obtained after 10 random repeated runs.
  • Figure 3: Simple regret vs iterations for battery manufacturing optimization experiments: (a) Optimization of calendering process (b) Optimization of battery endurance.
  • Figure 4: A complete process flowchart of our proposed BOAP framework.
  • Figure 5: Components of Arm$\mathfrak{-f}$ and Arm$-\mathfrak{h}$ in BOAP framework. Nodes highlighted in blue color corresponds to inputs or outputs of a Gaussian process. The estimated parameters and the latent variables are highlighted in yellow and green color, respectively. Rectangular boxes shaded in Grey and Orange correspond to the nodes representing rank GPs and conventional GPs, respectively.
  • ...and 1 more figures

Theorems & Definitions (3)

  • Definition 1: $\epsilon$-independence
  • Definition 2: Eluder dimension
  • Definition 3