Internships Mid-Atlantic 2021

---

Organization: IFM

Intern: Maria Deliyianni and Randy Price (University of Maryland Baltimore County)

An ultrasonic flow meter measures the time of flight (ToF), i.e. the time taken by a signal to travel a distance through a medium, to estimate the velocity of the liquid. Unfortunately, the accuracy of the ToF estimates can be affected by multiple factors, including noise and signal bandwidth, which produces outliers in the data. In this project we tested multiple approaches to eliminate the presence of outliers, including the cross correlation principle, the generalized cross correlation, and the application of various filtering techniques. Finally, an alternative approach was employed which combines the cross correlation principle along with the direct calculation of the time difference between upstream and downstream signals.

Organization: Society for Industrial and Applied  Mathematics

Intern: Eammon Hart (Drexel University)

Organization: Combustion Science and Engineering

Intern: Emily Kelting (Drexel University)

This project studies the dynamic behavior of combustive systems via bifurcation analysis. To conduct the computations, a numerical continuation software, as well as a chemical interpreter, were needed. A Linux machine was built with the softwares AUTO-07p and Cantera to perform the bifurcation analysis and mechanism interpretation, respectively. This report discusses the key steps it took to create that system. Additionally, utilizing the research of Lengyel and West, an example of a methane partial oxidation in a Perfectly Stirred Reactor is shown. Its thermal ignition and extinction points, steady-states, and one-parameter bifurcation curves are identified.

Organization: Intelligent Automation

Intern: Melinda Kleczynski (University of Delaware)

Multiple disciplines increasingly collect high-dimensional data. It can be difficult to visualize this data and discover useful features and relationships. There might be some low-dimensional structure, such as a manifold approximating the data. A useful tool for exploring this type of data is geometric multi-resolution analysis (GMRA). GMRA combines nested partitions of the data with local affine approximations at a range of scales. This project applied GMRA to both low-dimensional artificial datasets and high-dimensional real-world datasets to better understand how to leverage GMRA as an analysis tool. We focused on vector representations of protein sequences as a source of high-dimensional data. Our focus was on understanding and visualizing the output of GMRA. Primarily using Python, we considered dimensionality estimation, how categorical protein properties are organized in the datasets, and relationships with numerical protein properties. In particular, using protein sequences from the thermophile Aquifex aeolicus, we found that GMRA helped illustrate trends in properties such as the aliphatic index. This suggests future work to further investigate the real-world interpretation of vector representations of protein sequences

Organization: Dynamhex

Intern: Aditya Kumar (Johns Hopkins University)

In this project we tackled a few problems that arise from issues related to climate change. Specifically, given a private entity interested in fighting climate change with some set climate goals, what should be the course of action on a yearly, monthly and daily basis to meet those goals? How can they ensure that they meet them within their fiscal constraints? By finding solutions to these questions we contribute in a small way towards fighting climate change efficiently.

Organization: Intelligent Automation

Intern: Axel La Salle (Arizona State University)

Data routing, transformation, guaranteed delivery, and provenance are some of the critical components of systems of systems infrastructures. NiFi's feature set provides a powerful and reliable system to process and distribute data. Having the ability to manage a group of data in a specific set, called parameter contexts, that's available globally within the instance is a desired property for developers. In project 1, the management of parameter context is explored and the various ways to control access to them are highlighted.

Organization: Tanzen Medical

Intern: Dominic Macaluso (Drexel University)

Stress is an unavoidable part of day-to-day life. Stress can initiate bio-processes that alter human behavior; in many cases, stress leads to undesirable outcomes, such as poor performance. There are things that we can do to manage stress, such as meditation, time off from work, etc. However, many times, by the time stress management techniques are implemented, performance has already drastically decreased. As a result, it becomes desirable to determine a method for monitoring measurable bio-markers of excess stress. In this project, we seek a correlation between leg movements during sleep (LM) and daytime stress via heart rate variability (HRV) metrics. By analyzing subject data for LM and HRV, we can determine if an association exists between these two variables, giving users the ability to detect when stress will cause a decrease in performance. The results of the study are still being investigated. 

Organization: Johns Hopkins Centers for Civic Impact

Intern: Luan Nguyen (University of Maryland Baltimore County)

Organization: Emerald Development Managers

Intern: Yashil Sukurdeep (Johns Hopkins University

Organization: Johns Hopkins SNF Agora Institute

Intern: Nikki Wang (Johns Hopkins University

We investigate the grants from all private foundations in 2019. More precisely, we focus on the grants distribution, the giving strategies, the impact to recipients, etc. Moreover, we concentrate on the largest foundation and its donations over five years and hope to predict the donation for future years based on the existed grant-making. We study the change in total grants, number of large grants, the type of grant recipients. Then, using modeling techniques to maximize the foundation's utility for future grants.

Organization: Johns Hopkins University Malone Center for Engineering in Medicine

Intern: Boyang Xu (University of Delaware)

Surgical videos can provide a wealth of information about the performance of surgery. The rating process has been hitherto done manually. It is time-consuming and takes up lots of medical resources, since it needs residents or experienced surgeons to make it convincing. Therefore, an automated rating system is necessary. A single surgical video could be cut into several pieces, because of the limitation of storage. In the first part, a software, developed using Python and FFmpeg, for processing videos of surgical procedures for analysis of surgical performance is described. When concatenated videos are available, several deep learning neural network architectures are built in PyTorch, and trained, validated and tested to rate the performance of surgery.

Organization: Qualytics

Intern: Mingkai Yu (University of Maryland Baltimore County)

Data quality problems affect any company and government agency that uses data to make decisions. Qualytics aims to elevate trust and confidence in decision making by improving data accuracy. At Qualytics, we build a platform that identifies erroneous data at the source and prevents it from entering downstream data analysis pipelines. In this project, we attempt to apply novel statistical and machine learning methods to better catch the anomalies. One approach for examining the data is through normality tests, that is, to check if data roughly follows a normal distribution. We find normalcy tests are helpful indicators of the health of the data, though they have a tendency to be overly strict for many real-world large datasets. On another aspect, we introduce time-series analysis approaches to better understand historical data and dynamically estimate an appropriate range of data. We test different methods under several configurations and find a linear model is more suitable for most situations. This project as part of the data science initiative throws light on further exploration of the measurement and enhancement of data quality.