<p dir="ltr">Chemical reaction networks govern polymer degradation and contain critical design information regarding specific susceptibilities, degradation pathways, and degradants. However, predicting reaction pathways and characterizing complete reaction networks has been hindered by high computational costs because of the vast number of possible reactions at deeper levels of network exploration. In the first section, an exploration policy based on Dijkstra's algorithm on YARP using the reaction rate as a cost function was shown to provide a tractable means of exploring the pyrolytic degradation network of a representative commodity polymer, PEG. The resulting network is the largest reported to date for this system and includes pathways out to all degradants observed in earlier mass spectrometry studies. The initial degradation pathway predictions were validated by complementary experimental analysis of pyrolyzed PEG samples by ESI-MS. These findings demonstrate that reaction network characterization is reaching sufficient maturity to be used as an exploratory tool for investigating materials degradation and interpreting experimental degradation studies.</p>

<p dir="ltr">Modern machine learning research has recently made impressive progress across various research disciplines, such as computer vision, natural language processing, also in scientific fields including materials and molecule discovery, chip, and circuit design. In photonics/optics area, conventional methods in designing and optimiza- tion typically demand substantial time and extensive computing resources, where machine learning approaches hold the potential to significantly elevate and expe- dite these processes. On the other hand, machine learning algorithms can benefit from optical/photonics based neuromorphic computing systems due to their unique strengths in power consumption and parallelization. This talk will focus on imple- menting machine learning algorithms to optimize the optical/ photonics device (ML for photonics) as well as building optical based computing system for ML applica- tions (photonics for ML): First, I will discuss my work using probabilistic generative model (CVAE) for designing nanopatterned photonics power splitter with arbitrage splitting ratio. The model is incorporated with adversarial censoring and active learn- ing to increase the quality of generated devices. Next, I will report a physics-guided and physics-explainable recurrent neural network for time dynamics discovery in op- tical resonances, which can precisely forecast the time-domain response of resonance features with a very short portion of the initial input. The model is trained in a two-step multi-fidelity framework for high-accuracy forecast. In the end, I will present our progress in developing free space reconfigurable optical computing sys- tems for scientific computing, which is an optical based general matrix multiplication (GEMM) hardware accelerator by engineering a spatially reconfigurable array made from chalcogenide phase change materials. A device-system co-design methodology was implemented for GEMM system optimization. The device has been demonstrated over a various of ML applications.</p>

<p dir="ltr">Uncertainty quantification is a science and engineering subject that aims to quantify and analyze the uncertainty arising from mathematical models, simulations, and measurement data. An uncertainty quantification analysis usually consists of conducting experiments to collect data, creating and calibrating mathematical models, predicting through numerical simulation, making decisions using predictive results, and comparing the model prediction with new experimental data.</p><p dir="ltr">The overarching goal of uncertainty quantification is to determine how likely some quantities in this analysis are if some other information is not exactly known and ultimately facilitate decision-making. This dissertation delivers a complete package, including theory, algorithms, and applications of information field theory, a Bayesian uncertainty quantification tool that leverages the state-of-the-art machine learning framework to accelerate solving the classical uncertainty quantification problems specified by differential equations.</p>

<p>Current watershed-scale, nonpoint source (NPS) pollution models do not represent the processes and impacts of agricultural best management practices (BMP) on water quality with sufficient detail. To begin addressing this gap, a novel process-based, watershed-scale, water quality model (WEPP-WQ) was developed based on the Water Erosion Prediction Project (WEPP) and the Soil and Water Assessment Tool (SWAT) models. The proposed model was validated at both hillslope and watershed scales for runoff, sediment, and both soluble and particulate forms of nitrogen and phosphorus. WEPP-WQ is now one of only two models which simulates BMP impacts on water quality in ‘high’ detail, and it is the only one not based on USLE sediment predictions. Model validations indicated that particulate nutrient predictions were better than soluble nutrient predictions for both nitrogen and phosphorus. Predictions of uniform conditions outperformed nonuniform conditions, and calibrated model simulations performed better than uncalibrated model simulations. Applications of these kinds of models in real-world, historical simulations are often limited by a lack of field-scale agricultural management inputs. Therefore, a prototype tool was developed to derive management inputs for hydrologic models from remotely sensed imagery at field-scale resolution. At present, only predictions of crop, cover crop, and tillage practice inference are supported and were validated at annual and average annual time intervals based on data availability for the various management endpoints. Extraction model training and validation were substantially limited by relatively small field areas in the observed management dataset. Both of these efforts contribute to computational modeling research and applications pertaining to agricultural systems and their impacts on the environment.</p>