Current Research Projects

  1. Application of Dueling Networks to UAV (Supported by US Air Force).
    We will apply the dueling networks to UAV's path from the origin to the target. On the path there are some obstacles and they have to learn how to navigate the shortest path between them.
  2. Navy Oil Analysis and Lubricants Programs, Statistics and Data Analytics Studies (Supported by Naval Air Systems Command (NAVAIR)).
    We will use machine learning models to develop techniques to characterize the mechanical condition of fleet aircraft and development of wear metal results limits.
  3. Classification of Incomplete Networks (Supported by ONI via NRP).
    The rise of accessible real-world data creates a growing interest in effective methods for accurate network classification, especially for networks with incomplete information. The intelligence community requires an understanding of a network before the team can develop a strategy to combat the adversary. Problems are typically time-sensitive; however, gathering complete and actionable intelligence is a challenging mission. An adversary’s actions are secretive in nature. Crucial information is deliberately concealed. Intentionally dubious information creates problematic noise. Therefore, if an observed incomplete network can be classified as-is without delay, the network can be properly analyzed for a strategy to be devised and acted upon earlier. This project considers a method for classification of incomplete networks. We examine the effects of training the classification model with complete and incomplete information. Observed network data and their graph features are classified into technological, social, information, and biological categories using supervised learning methods.
  4. Inference on Missing Information on a Social Network (Supported by N1 via NRP).
    Networks and graphs have long been a subject of study, but with an explosion in the amount of available data to describe them, machine learning (ML) methods have become a popular compliment to traditional network analysis techniques. This is particularly true when the challenge of uncertainty enters the picture, but can be overcome with the application of ML methods with large amounts of data. Understanding a social network between workers can be used for modeling social relationship factorial analysis to preventing ore reducing harassments among them. In order to analyze as accurate as possible, it is important to have an accurate social network which a model will be based on and we assume that the observed social network is built with a complete information. However, often a victim of a sexual harassment or a work harassment never report their relations with their attackers. In a reality an observed social network is very often build with missing information. We propose here to infer connectedness of the social networks in a community within the Navy from a missing information. After correctly selecting a model to infer missing part of the social network we will analyze strength of relationships between workers. In this social network we set workers in a group as nodes and we draw edge between nodes if they have some social interaction between them. By this way we can construct several social networks, like communication networks. Then based on the social network we reconstruct we will conduct logistic regressions to see which factors contributing to each relationship.
  5. Principal Component Analysis over tree spaces and its applications to phylogenomics (Supported by NSF).
    Phylogenomics is a relatively new field that seeks to understand evolutionary relationships between organisms at the scale of the whole genome. One of the central questions in evolutionary biology is a better understanding of the relationships between organisms, usually summarized in the form of a phylogenetic tree. The methods in common use for developing these trees tend to work best for closely related organisms, and when the sequences are relatively short; for example, the DNA sequence for a single gene applied to a collection of mammals. When comparing more distantly related organisms, or data from large portions of the genome, current techniques can break down. Since modern technology can quickly and cheaply produce genome-scale sequence data, there is a pressing need for better analytical tools tailored to this large-scale high-dimensional data. The most popular statistical methods for finding general patterns in large-scale data, such as Principal Component Analysis (PCA), make the assumption that the space where the data lies is flat, like the plane geometry of Euclid. However, the space of possible phylogenetic trees has a decidedly non-Euclidean geometry, with a surface more akin to an origami figure made with a sheet of rubber. The goal of this project is to develop alternative types of principal components, and methods to calculate them, which take into account the unusual structural features of the mathematical space of phylogenetic trees.
  6. Developing tropical support vector machines over the tropical projective space and treespaces.