Research
Research Overview
My research lies at the intersection of Mathematics and Artificial Intelligence. I develop efficient, theoretically grounded algorithms augmented by machine learning and neural networks for analyzing high-dimensional data that are often noisy, incomplete or corrupted. These methods have broad impact across hyperspectral imaging, video surveillance, biomedical data, and Internet of Things (IoT) systems.
Robust Matrix Completion
A core area of my work is Robust Matrix Completion, an approach designed to simultaneously handle incomplete data and sparse outliers. I have developed algorithms such as Robust CUR Completion (RCURC) and Learned Robust Matrix Completion (LRMC), a deep learning-augmented framework that demonstrate strong performance across multiple domains:
- Video Surveillance: Video background subtraction.
- Imaging: Ultrasound imaging and face modeling.
- Remote Sensing: Cloud removal from multi-temporal satellite imagery.
Video 1: Background subtraction using LRMC
Distance Geometry Problems
I also focus on Distance Geometry problems, which involve finding the point configurations from the distances measured among those points. I’ve developed algorithms like Structured Robust Euclidean Distance Geometry (SREDG) and Robust Euclidean Distance Geometry via Dual Basis (RoDEoDB), which show strong results in critical applications:
- Localization: Sensor localization.
- Bioinformatics: Protein structure prediction (molecular conformation).

Figure 1: Visual representation of reconstruction of Protein 1AX8 by RoDEoDB under different corruption levels. The reconstructed protein structure (in brown) closely aligns with the true protein structure (in blue).
Weighted Matrix Completion
Weighted Matrix Completion (WMC) is another key area of my research, aimed at incorporating prior subspace information to improve recovery performance. One primary application is compressive hyperspectral imaging in pushbroom remote sensing. Hyperspectral data are acquired as a 3D cube of spatial pixels $(x, y)$ and spectral channels $(z)$, as shown in Figure 2. The central challenge is to reconstruct each high-dimensional data slice $D$ in real time from only a small, incomplete set of observed entries $b$.

Figure 2: From satellite acquisition to ground reconstruction: DMD-based compression produces sparse measurements $b$, then weighted matrix completion recovers the hyperspectral slice $D$.
We are currently developing a streaming WMC framework that leverages information from the previously reconstructed slice to guide recovery of the current slice. In addition, we use machine learning to adaptively learn slice-specific weights, enabling fast and accurate real-time reconstruction.
Collaborative Research
As a Data Scientist, my expertise extends to interdisciplinary research where I apply data-driven methods to challenges beyond core algorithms. I have worked with colleagues at the Nanoscience and Technology Center, UCF, applying machine learning to predict the discharge capacity of $\text{MnO}_2$ cathodes for high-performance zinc-ion batteries. I have also contributed to digital humanities, using statistical techniques for authorship attribution. I am eager to continue pursuing research that bridges disciplines and leverages my strengths in data science.
Research & Publication
-
Cai, HanQin; Kundu, Chandra; Liu, Jialin; Yin, Wotao (2026). Deeply Learned Robust Matrix Completion for Large-scale Low-rank Data Recovery. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI). arXiv. (Equal contribution)
-
Yengejeh, Amir Alipour; Tang, Larry; Bridge, Candice; Kundu, Chandra (2026). Exploring Scientific Literature Using Topic Modeling: A Practical Framework for Discovery and Classification. Informatics.
-
Kundu, Chandra; Tasissa, Abiy; Cai, HanQin (2025). Structured Sampling for Robust Euclidean Distance Geometry. 59th Annual Conference on Information Sciences and Systems (CISS 2025). arXiv.
-
Kundu, Chandra; Tasissa, Abiy; Cai, HanQin (2025). A Dual Basis Approach for Structured Robust Euclidean Distance Geometry. arXiv preprint arXiv:2505.18414.
-
Chowdhury, N. A.; et al. (2025). Machine learning based prediction of MnO2 cathode discharge capacity for high-performance zinc-ion batteries. Physical Chemistry Chemical Physics (Royal Society of Chemistry).
-
Cai, HanQin; Huang, Longxiu; Kundu, Chandra; Su, Bowen (2024). On the Robustness of Cross-Concentrated Sampling for Matrix Completion. 58th Annual Conference on Information Sciences and Systems (CISS 2024). arXiv.
-
Zheng, L.; Zheng, H.; Kundu, Chandra (2023). Authorship Attribution via Occupancy Problem Type Indices. Journal of Quantitative Linguistics.
-
Kundu, Chandra (2020). Book Genre Classification By Its Cover Using A Multi-view Learning Approach. M.S. thesis, Western Kentucky University.
Presentations and Talks
-
UCF Statistics & Data Science Colloquium, Orlando, Florida.
Talk: Learning Low-Rank Structure in Imaging and Geometry (with Dr. HanQin Cai). (April 2026) -
Foundations of AI Seminar, University of Central Florida, Orlando, Florida.
Talk: Data-Driven Algorithms for Generalized Low-Rank Matrix Recovery. (March 2026) -
SIAM Conference on Mathematical & Computational Issues in the Geosciences (GS25) in Baton Rouge, Louisiana. Co-Author Presentation: Weighted Matrix Completion: Applications to Compressive Hyperspectral Imaging (Presented by Oscar Lopez). (October 2025)
-
59th Annual Conference on Information Sciences and Systems (CISS 2025), Baltimore, Maryland.
Talk: Structured Sampling for Robust Euclidean Distance Geometry. (March, 2025) -
Annual ASA Florida Chapter Meeting, Orlando, Florida.
Poster Presentation: Deep Learning-Augmented Robust Matrix Completion for Cloud Removal in Satellite Imagery. (February, 2025) -
SIAM Conference on Mathematics of Data Science (MDS24), Atlanta, Georgia.
Poster Presentation: Learned Large-Scale Robust Matrix Completion Via Deep Unfolding - Application to Cloud Removal from Multi-Temporal Satellite Imagery. (October, 2024) -
SIAM Conference on Imaging Science (IS24), Atlanta, Georgia.
Poster Presentation: Learned Large-Scale Robust Matrix Completion Via Deep Unfolding - Application to Video Background Subtraction. (May, 2024) -
58th Annual Conference on Information Sciences and Systems (CISS 2024), Princeton, New Jersey.
Co-Author Presentation: On the Robustness of Cross-Concentrated Sampling for Matrix Completion (Presented by Dr. HanQin Cai). (March, 2024) -
Math Graduate Student Seminar, Western Kentucky University.
Invited Talk: A Trainable Approach to Large-Scale Robust Matrix Completion. (February, 2024) -
Algorithms for Threat Detection (ATD) PI Workshop, Fairfax, Virginia.
Co-Author Presentation: Geopolitical Event Forecasting: A Quantile-Based Approach Integrating Gaussian Process Regression and Temporal Fusion Transformer (Presented by Hayden Hampton). (October 2023) -
39th Annual Mathematics Symposium, Western Kentucky University.
Talk: Authorship Attribution via Occupancy Problem Type Indices. (November, 2019)
Reviewer Service
- International Conference on Big Data, IoT and Machine Learning (BIM) (2023)
Funding & Support
- Nonconvex Models for Structured Sensing: Theory, Algorithms, and Applications, NSF. ($121,425)
- Role: Research Assistant
- PI: HanQin Cai