We use AI and UQ to accelerate scientific discoveries and improve decision-making.
Modeling the "Why" in
Business Operations
Business Operations
Businesses collect vast amounts of data, which they leverage with advanced modeling techniques to develop actions to enhance their operations. The goal is to create an AI engine that provides managers with actionable recommendations to produce a desired effect on key performance indicators.
Lowe's Innovation Fund
Causal Modeling of
Affective Polarization
Affective Polarization
The objective of this project is the development of an interactive causal modeling framework to understand affective polarization. The focus is on bridging the gap between data-driven methods for discovering causal relations and expert domain knowledge to mitigate the impact of data limitations.
ARO W911NF-22-1-0035
Deep Learning for
Surface Chemistry
Surface Chemistry
In this project, we are using computational catalysis and deep learning to improve the analysis and understanding of surface chemistry. We will use large datasets of DFT energies and advanced neural networks to identify invariant material and molecular descriptors that can help us predict the behavior of chemical systems.
NSF 2218841
From Causal Pairs to Causal Graphs
The paper proposes new methods for causal structure learning from observational data using probability distributions informed by cause-effect pair features. These methods are compared to traditional and state-of-the-art approaches on synthetic and real datasets, and have similar or better performance while being computationally faster.
Expert Knowledge in Causal Discovery
Comparative analyses show that knowledge that corrects mistakes in the model can lead to significant improvements and that constraints on active edges have a larger positive impact than inactive edges. However, the induced knowledge does not correct a higher number of incorrect edges than expected on average.
Invariant Molecular Representations
The study introduces a machine learning method with Siamese neural networks to efficiently predict adsorption energies, using data from multiple density functional theory (DFT) functionals, significantly enhancing accuracy and robustness in catalyst screening, especially for propane dehydrogenation on platinum catalysts.