Marija Iloska

Our interest in obtaining predictive models from vast volumes of data has never been greater. Often, we want these models to be as simple as possible but not simpler. Reducing the complexity or size of a model is accomplished by enforcing sparsity in the model’s parameters or structure. In the context of sparse regression, LASSO and its variants have become standard practice, but they rely on the choice of the penalty parameter for desired degrees of sparsity or predictability. In this work, we develop a novel online inference approach for transdimensional model learning with online feature selection, built on the fundamental principles of predictive least squares. The solution path follows the predictive error at each step, prioritizing more predictive features in the model. We provide a variety of examples to analyze the capabilities of the proposed method and evaluate its performance against both standard and recently proposed feature selection methods.

We demonstrate a data-driven approach to interpreting surface reactions by combining time-resolved gas pulsing infrared spectroscopy with chemical reaction neural networks (CRNNs). Using CO adsorption and desorption on Pd(111) at 460-490 K as a model system, we show how transient kinetic data can reveal detailed reaction mechanisms. Starting with a simple one-species model, we systematically evaluate increasingly complex mechanisms involving hollow and bridge site adsorption. Despite the similar goodness of fit to the same experimental absorbance data, our models predict distinct coverage dynamics for different adsorption sites. Through analysis of spectral peak stability and predicted dynamics, we identify a mechanism in which CO primarily adsorbs on bridge sites followed by rapid conversion to hollow sites as being the most physically consistent with experimental observations. This work provides a framework for extracting mechanistic insights from limited experimental data, demonstrating how machine learning can bridge the gap between transient kinetic measurements and a molecular-level understanding of surface reactions.

In this work, we develop a novel Bayesian approach to study the adsorption and desorption of CO onto a Pd(111) surface, a process of great importance in natural sciences. The motivation for this work comes from the recent availability of time-resolved infrared spectroscopy data and the need for model interpretability and uncertainty quantification in chemical processes. The objective is to learn the relevant parameters that characterize the process: coverage with time, rate constants, activation energies, and pre-exponential factors. Our approach consists of three main schemes: (i) a problem design and probabilistic model for the whole system, (ii) a particle Markov chain Monte Carlo sampler to learn the hidden coverages and rate constant parameters, and (iii) two Bayesian formulations to infer the activation energies and pre-exponential factors. The flexibility of the Bayesian framework allows for uncertainty quantification where possible and integration of mathematical constraints in the model to reflect the system physically. We found that our results for the activation energies and pre-exponential factor are in agreement with those reported in the experimental literature, independently, and we provide discussions on the advantages and disadvantages as well as applicability to other systems.

We demonstrate a data-driven approach to interpret surface reactions by combining time-resolved gas-pulsing infrared spectroscopy with Chemical Reaction Neural Networks (CRNN). Using CO adsorption and desorption on Pd(111) at 460K-490K as a model system, we show how transient kinetic data can reveal detailed reaction mechanisms. Starting with a simple one-species model, we systematically evaluate increasingly complex mechanisms involving hollow- and bridge-site adsorption. Despite similar goodness of fit to the same experimental absorbance data, our models predict distinct coverage dynamics for different adsorption sites. Through analysis of spectral peak stability and predicted dynamics, we identify a mechanism where CO primarily adsorbs on bridge sites followed by rapid conversion to hollow sites as the most physically consistent with experimental observations. This work provides a framework for extracting mechanistic insights from limited experimental data, demonstrating how machine learning can bridge the gap between transient kinetic measurements and molecular-level understanding of surface reactions.

An important and often overlooked aspect of particle filtering methods is the estimation of unknown static parameters. A simple approach for addressing this problem is to augment the unknown static parameters as auxiliary states that are jointly estimated with the time-varying parameters of interest. This can be impractical, especially when the system of interest is high-dimensional. Multiple particle filtering (MPF) methods were introduced to try to overcome the curse of dimensionality by using a divide and conquer approach, where the vector of unknowns is partitioned into a set of subvectors, each estimated by a separate particle filter. Each particle filter weighs its own particles by using predictions and estimates communicated from the other filters. Currently, there is no principled way to implement MPF methods where the particle filters share unknown parameters or states. In this work, we propose a fusion strategy to allow for the sharing of unknown static parameters in the MPF setting. Specifically, we study the systems which are separable in states and observations. It is proved that optimal Bayesian fusion can be obtained for state-space models with non-interacting states and observations. Simulations are performed to show that MPF with fusion strategy can provide more accurate estimates within fewer time steps comparing to existing algorithms.