Domain Knowledge Guided Bayesian Optimization For Autonomous Alignment Of Complex Scientific Instruments

TL;DR

High-dimensional, tightly coupled control spaces in complex scientific instruments create needle-in-a-haystack optimization challenges for standard Bayesian methods. The authors introduce domain knowledge guided Bayesian Optimization, combining physics-informed coordinate transformations to align active subspaces with optimization axes and a reverse annealing schedule to sustain exploration, demonstrated on the 12-parameter HXRSND benchmark. The approach yields a two-component framework (transformation plus reverse annealing) that reliably reaches the global optimum where BO, TurBO, and MOBO struggle, and it offers a generalizable blueprint for domain-guided optimization in other complex physics systems. This method promises improved sample efficiency, robustness, and scalable autonomous tuning for large-scale scientific facilities.

Abstract

Bayesian Optimization (BO) is a powerful tool for optimizing complex non-linear systems. However, its performance degrades in high-dimensional problems with tightly coupled parameters and highly asymmetric objective landscapes, where rewards are sparse. In such needle-in-a-haystack scenarios, even advanced methods like trust-region BO (TurBO) often lead to unsatisfactory results. We propose a domain knowledge guided Bayesian Optimization approach, which leverages physical insight to fundamentally simplify the search problem by transforming coordinates to decouple input features and align the active subspaces with the primary search axes. We demonstrate this approach's efficacy on a challenging 12-dimensional, 6-crystal Split-and-Delay optical system, where conventional approaches, including standard BO, TuRBO and multi-objective BO, consistently led to unsatisfactory results. When combined with an reverse annealing exploration strategy, this approach reliably converges to the global optimum. The coordinate transformation itself is the key to this success, significantly accelerating the search by aligning input co-ordinate axes with the problem's active subspaces. As increasingly complex scientific instruments, from large telescopes to new spectrometers at X-ray Free Electron Lasers are deployed, the demand for robust high-dimensional optimization grows. Our results demonstrate a generalizable paradigm: leveraging physical insight to transform high-dimensional, coupled optimization problems into simpler representations can enable rapid and robust automated tuning for consistent high performance while still retaining current optimization algorithms.

Figures (9)

• Figure 1: Schematic of the HXRSND with the CC (Channel Cut) branch in blue and the delay branch in red. The x-ray beam propagates from right to left. The arrows indicate the motorized degrees of the system. In addition, each crystal along the delay branch has "chi" adjustment which corresponds to rotation about the tangential vector of the crystal surface (not shown). The red and blue dots correspond to locations of beam diagnostics, as noted in the legend on the top right of the figure Zhu2017.
• Figure 2: Schematic outlining the key challenges in the optimization of the HXRSND, (a) Contour surfaces of the beam intensity in a 2D subsection of the 8D space exhibiting the sparsity of the optimum and the lack of informative gradients, (b) Contour surfaces of the beam position error exhibiting the interdependence between input features. For both (a) and (b), the inputs correspond to the t1.th1 and t1.th2 settings varying over $\pm 100$ microrads (see Fig. \ref{['fig:fig1']}).
• Figure 3: Schematic outlining (a) the interdependence between inputs in the Beam Position Error objective in $\mu$m (microns) (b) the sparsity of the Beam Intensity objective over the input space in arbitrary units (a.u.). For both (a) and (b), the inputs perturbations are marked in units of microrads.
• Figure 4: Comparison between the performance of domain knowledge guided Bayesian Optimization (labeled as Domain Guided), against standard Bayesian Optimization (BO), Trust Region based Bayesian Optimization (TurBO) and Multi-Objective Bayesian Optimization (MOBO) for the Beam Position Error. For each algorithm, we report the median of the running minimum with the solid line, over $25$ experiments. The shaded zones outline the $25^{th}$ and $75^{th}$ percentiles.
• Figure 5: Comparison between the performance of domain knowledge guided Bayesian Optimization (labeled as Domain Guided), against standard Bayesian Optimization (BO), Trust Region based Bayesian Optimization (TurBO) and Multi-Objective Bayesian Optimization (MOBO) for the Beam Intensity. For each algorithm, we report the median of the running maximum with the solid line, over $25$ experiments. The shaded zones outline the $25^{th}$ and $75^{th}$ percentiles.