Abstract (click to read)

In many transit systems including the Duetsche Bahn Railway in Europe, passengers are legally required to purchase a ticket before riding the train, but are not physically forced to do so. Instead, train inspectors are deployed throughout the transit system to inspect passengers, and fine people who are riding the train illegally. With approximately 7 million people taking the train every day, it is crucial that inspectors are deployed in a manner that will maximize the number of passengers that are controlled.

This summer I had the opportunity to travel to Germany for the GRIPS program offered through IPAM, in order to try and model/solve this problem for the Deutsche Bahn Railway System. In this talk, I will discuss how my team and I approached the problem, and some of the preliminary results we obtained. I will also discuss my experiences with the IPAM GRIPS program, in case anyone is interested in participating in it next summer.

Abstract (click to read)

Quantum information science (QIS) is an emerging field with an immense potential to revolutionize various science and engineering applications involving quantum computing, communication, and quantum simulation. The talk will provide an overview of the QIS area, including its key applications, challenges, and opportunities, as well as key quantum related activities happening at PNNL.

We will join the lecture from Pullman via Zoom.

Abstract (click to read)

We consider new approaches to characterize the robustness of solutions to a system of nonlinear equations. This problem arises in many applications such as the power grid and other infrastructure networks. We use techniques from algebraic topology (topological degree theory) to characterize the robustness margin of such systems of equations. We then cast the problem of checking for the specified conditions as a nonlinear optimization problem. Based on this formulation, we develop efficient computational techniques to estimate lower and upper bounds for the robustness margin.

A preprint is available on arXiv:1907.12206.

Abstract (click to read)

In order to derive unbiased inference from observational data, matching methods are often applied to produce balanced treatment groups in terms of relevant background variables. Although many matching algorithms exit in the literature, most require a large control reservoir and can not deal with missing data. Random forest, averaging outcomes from many decision trees, is non-parametric in nature, can deal with missing data in the tree building process, and can produce more accurate and less model dependent estimates of propensity scores as well as a proximity matrix. In this study, iterative matching algorithms are developed in order to form balanced samples based on limited sample sizes for both groups. A R package implementing the proposed methods has also been developed. The proposed methods are applied to two data sets, arising from studies of autism spectrum disorder (ASD) and student success.

We will join by Zoom.

Abstract (click to read)

I will revisit the field classical mathematical epidemiology starting from the contributions of Sir Ronald Ross in 1911. Extensions of Ross’ framework and recent applications will be discussed in the context of the study of the spread of influenza in the presence of cross-immunity as well disease dynamics in models that include social dynamics, in particular on those that incorporate the role of individual decisions. Examples will be provided from communicable diseases under scenarios that account for various modes of transmission.

We will join by Zoom.

Abstract (click to read)

The aim of this talk will be to try and understand the UMAP algorithm (Uniform Manifold Approximation and Projection) from a "user's" perspective. Most of our effort will be spent trying to understand how the algorithm works intuitively, rather than understand the mathematical machinery underpinning the algorithm.

Abstract (click to read)

Biological signaling pathways describe a series of molecular reactions that are initiated by an external signal and culminate in a specific cellular response (for example, a cell may grow, divide, move, or die). The wealth of publicly-available data containing signaling reactions provide an opportunity for automatically reconstructing such pathways using computational approaches. In this talk, I will describe network-based methods that automatically reconstruct signaling pathways from protein-protein interaction data. Our methods successfully reconstruct human signaling pathways with much higher precision and recall than several state-of-the-art graph algorithms. Further, our methods suggest experimentally testable hypotheses about canonical signaling. Extensions of this work have constrained pathway reconstructions by considering the spatial localization of interactions within the cell, and I will also describe our first steps in using ideas from pathway reconstruction to discover signaling pathways that are altered in cancer.

Abstract (click to read)

A common framework for describing the function of auditory cortical neurons is the linear-nonlinear spectro-temporal receptive field (LN STRF). This model casts the time-varying neural spike rate as a linear weighted sum of the immediately preceding sound spectrogram, followed by nonlinear rectification. Two models have expanded on the LN STRF to account for sensory contextual effects, using short-term plasticity (STP) or contrast-dependent gain control (GC). Both improve performance over the LN model, but they have never been compared directly. Thus, it is unclear whether they account for distinct processes or describe the same phenomenon in different ways. To address this question, we recorded the activity of single primary auditory cortical neurons (n = 423) in awake ferrets (n = 7) during presentation of natural sound stimuli. We fit STRFs incorporating one nonlinear mechanism (GC or STP) or both mechanisms (GC+STP) on this single dataset. We compared model performance using the correlation coefficient (Pearson's R) between predicted and observed time-varying spike rate for each neuron. Our results indicate that the STP model performs significantly better than the GC model, and that the GC+STP model performs significantly better than either individual model. This finding indicates that the STP and GC models explain distinct aspects of neural activity. Further, the success of the combined model suggests that auditory cortical neurons utilize at least two independent mechanisms to adapt encoding properties to different sensory contexts. Future neuromorphic sound processing technologies may therefore improve their performance by incorporating both STP- and GC-based computations.

This is joint work with Dr. Stephen David, OHSU.

Abstract (click to read)

In chemical systems, water and organic liquids are typically separated by an interface surface. As part of the chemical reaction, water molecules escape to the other side through tube-like features called protrusions. Our goal is to objectively define which features are actually protrusions, and also identify and characterize them automatically. We develop a computational framework for characterizing protrusions on liquid-liquid interfaces using the concept of multiscale flat norm from geometric measure theory.

Joint work with the Aurora Clark group. Presentation will be delivered live from Pullman via video.

Abstract (click to read)

For survey stats practitioners, it sometimes feels like the sky is falling. Response rates are declining. Data collection costs are increasing. Federal budgets are shrinking. But, with the advent of big data and advancements in technology, a wealth of new data sources, such as satellite imagery, are available to supplement survey data. And, flexible, predictive models, such as regression trees and elastic net regression, can be an effective tool to combine these different data sources and to produce more efficient estimators. Drawing on my collaborations with the U.S. Bureau of Labor Statistics and the U.S. Forest Inventory and Analysis Program, we'll explore the utility of predictive models in survey estimation.

Abstract (click to read)

A persistent challenge in applied mathematics is balancing expressiveness in programs with computational performance. In an ideal world, we would express a program in the language of our mathematical domain and let the computer compute a performant program that implements it. Unfortunately, we often are forced to give up one of these: choose expressivity at the cost of performance, or performance at the cost of expressivity. It doesn't have to be this way. Deductive synthesis is a mechanism by which programs can be deduced from specifications through a refinement process. This talk will introduce the synthesis concept and discuss the technical challenges that have prevented it from being adopted in widespread use today. I will also introduce a testbed under development to make deductive synthesis accessible to users outside specific corners of computer science and formal mathematical logic.