Undergraduate Research in Computational Mathematics

Cutting-Edge Methods

Discover and explore modern applied mathematics. Apply new methods to interesting open problems.

Advanced Training

Work individually or in teams to develop new algorithms, hone soft skills, and enhance your scientific background.

Career Building Projects

Get hands-on experience in modern research under faculty guidence and explore opportunities in STEM.

Some Current Projects

Data-Driven Methods

Machine learning, data-driven methods, and optimization are useful tools for studying structures and trends in high dimensional datasets.

Numerical ODE and PDE

ODE and PDE are fundemental models in the physical sciences. Modern numerical solvers utilize solution behavior and data to increase efficiency and reduce complexity.

Image Recovery

Compression, noise, and data corruption are common issues in image processing. With optimization and PDE, rigourous models can be used to recover a suitable approximation. 

Previous Student Projects

Detecting Foot-Chases From Police Body-Worn Video

Students: Rafael Aguayo, Alejandro Camacho, Piyali Mukherjee, and Qi Yang

Existing methods to record interactions between the public and police officers are unable to capture the entirety of police-public interactions. In order to provide a comprehensive understanding of these interactions, the Los Angeles Police Department (LAPD) intends to utilize Body-Worn Video (BWV) collected from cameras fastened to their officers. BWV provides a novel means to collect fine-grained information about police-public interactions. The purpose of this project was to identify foot-chases from the videos using machine-learning algorithms. Our proposed algorithm used the Bag-of-Intrinsic-Words algorithm followed by classification via support-vector machines. Our training dataset consisted of 100 training videos (20 foot-chase & 80 non-foot-chase), and a test dataset of 60 LAPD videos (4 foot-chase & 56 non-foot-chase). We achieved results of 91.6% testing accuracy. In colloboration with Prof. P. Jeffrey Brantingham from UCLA.

Automated Detection of Chondrocytes in Growth Plate Images

Students: Emily Beylerian, Brian de Silva, Ben Gross, and Hannah Kastein

Analysis of chondrocyte development in bone growth plates yields data concerning foundational genetic structure. Previous research on chondrocyte organization and alignment has required manual detection of cell sizes and locations. This work proposed an algorithm for automated detection of chondrocytes within bone growth plate images. The method aimed to aid biological research by increasing the efficiency and consistency of collecting data from images of growth plate regions. Due to indistinct background cells and variations between regions of growth plates, classical techniques of clustering, segmentation, thresholding, and filtering fail to accurately identify cell boundaries. We proposed an automated algorithm that incorporates the methods of Retinex, anisotropic diffusion, and various thresholding methods in order to detect locations and sizes of chondrocytes. Our results and analysis demonstrated the effectiveness of the proposed method as applied to growth plate images. With Dr. Maria-Grazia Ascenzi.

Finding a Needle in a Haystack: An Image Processing Approach

Student: Emily Beylerian

Image segmentation (also known as object/edge detection) is the process of dividing an image into its constituent parts using information about the boundaries between objects, edges within objects, variations in intensity, etc. The human eye can easily recognize salient information from an image; however, background variations in intensity, noise, and other highly oscillatory features make the process of image segmentation computationally challenging. This work is unique because we propose using a cartoon-texture-noise separation to remove highly oscillatory features from the image prior to segmentation. A new numerical implementation is provided for one of the two decompositions used as well as various experimental results. The method is applied to the classic example of finding a needle in a haystack, as well as real images where the texture component and noise causes problems for standard techniques.

The Construction of a Vague Fuzzy Measure Through L1 Parameter Optimization

Students: Evgeni Dimitrov, Kizza Nandyose, Sandra Rankovic, and David Wen.

This work presented a method to construct an aggregation function, reflecting a complex set of initial user preferences, which can be used in the framework of multi-criteria decision making. We consider problems where the decision maker can provide information about the importance and interactions between criteria, as well as a desired portion of criteria to be satisfied. The proposed aggregation process is a vague Choquet integral, whose parameters are constructed in two steps. First, we solve a convex constrained L1 optimization problem to obtain a fuzzy measure reflecting the importances and interactions between the criteria. Then the measure is transformed by a monotonic mapping to include vague information on what portion of criteria has to be satisfied. The proposed approach provides an automated construction of an aggregation function, which is completely free of data learning and manual processing. In addition, this method provides a novel fuzzy measure that integrates two different classes of information: importance/interactions of criteria and vague statements. With Dr.  Olivier Thonnard.