The full text of this item is not available at this time because the student has placed this item under an embargo for a period of time. The Libraries are not authorized to provide a copy of this work during the embargo period, even for Texas A&M users with NetID.
Optimal Decision Making for Accelerating Scientific Discovery
Abstract
Scientific discovery is the process of finding answers to scientific inquiries. Scientific discovery
forms the basis of scientific/engineering applications as it serves as an operational objective or a means of achieving operational goals. In practice, scientific discovery is realized via (a sequence of) scientific decision-making that involves predicting the potential efficacy of available options and taking action that maximizes the expected utility of interest. Making optimal decisions is particularly important in real-world scientific/engineering applications as, potentially, it accelerates the discovery or even has a profound impact on the success of scientific applications.
In this dissertation, we comprehensively study the optimal decision-making problem for accelerating successful scientific discovery. Starting with a data-driven model that accelerates the optimal decision-making itself for a representative real-world scientific/engineering application, we propose mathematical optimization frameworks for identifying optimal decision-making. Based on the proposed models, we quantitatively analyze how fast this set of models can advance scientific discoveries for the applications.
In the first part, we consider the optimal decision-making problem in the context of optimal experimental design (OED). Identifying the optimal experiment that is expected to maximally reduce system uncertainty has become a critical problem in real-world scientific/engineering applications that involve modeling complex systems. Mean objective cost of uncertainty (MOCU)-based OED has shown that such a goal-driven OED is extremely useful in real-world problems that aim at achieving specific objectives based on complex uncertain systems. However, MOCU-based OED tends to be computationally expensive mainly due to the prediction cost of the potential efficacy of available experiments based on MOCU, which limits its practical applicability. To address this issue, we propose a novel ML scheme that can significantly accelerate MOCU computation and expedite MOCU-based experimental design. We apply the proposed ML-based OED acceleration scheme to design experiments aimed at optimally enhancing the control performance of uncertain Kuramoto oscillator models.
In the second part, we study an optimal decision-making problem for screening campaigns based on high-throughput virtual screening (HTVS) pipeline structures, which frequently arises in various scientific and engineering problems including drug discovery and materials design. We propose a general mathematical framework for optimizing HTVS pipelines that consist of multi-fidelity models. The central idea is to optimally allocate the computational resources to models with varying costs and accuracy to optimize the return-on-computational-investment (ROCI). Based on both simulated as well as real data, we demonstrate that the proposed optimal HTVS framework can significantly accelerate screening virtually without any degradation in terms of accuracy.
In the third part, based on the optimization framework we proposed in the second part of the dissertation, we design an optimal computational campaign (OCC) in the context of rapidly selecting redox-active organic materials for developing novel energy storage devices. Starting from a high-fidelity model that computes the redox potential (RP) of a given material, we show how a set of surrogate models with different accuracy and complexity may be designed to construct a highly accurate and efficient HTVS pipeline. Besides, we further generalize the screening condition and the optimization framework accordingly, which enables the design of computational screening campaigns according to a screening range. We demonstrate that the proposed HTVS pipeline remarkably enhances the overall throughput for a given computational budget.
Subject
Approximate mean objective cost of uncertainty (MOCU)high-throughput screening (HTS)
high-throughput virtual screening (HTVS)
Kuramoto model
machine learning (ML)
objective-based uncertainty quantification (objective-UQ)
optimal computational campaign
optimal experimental design (OED) acceleration
optimal screening
organic electrode materials
redox potential (RP)
renewable energy
return-on-computational-investment (ROCI)
Citation
Woo, Hyun-Myung (2022). Optimal Decision Making for Accelerating Scientific Discovery. Doctoral dissertation, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /197884.