Publications

You can also find my articles on my Google Scholar profile.

Hybrid Modeling of Harmful Algal Blooms in the Gulf of Maine Using Ocean Color and Regional Ocean Models

Published in Earth Arxiv, 2026

Harmful algal blooms (HABs) in coastal waters contaminate shellfish, close fisheries, and threaten public health, yet the physical and biological conditions that trigger a bloom are rarely captured at sufficient resolution by any single observing system. This paper develops a hybrid machine learning framework for species-level HAB detection in the Gulf of Maine, combining satellite ocean color, in-situ flow imaging, and regional ocean forecast model outputs.

Recommended citation: Rathore, K. J., Buckner, J. H., Watson, J. R. (2026). Hybrid Modeling of Harmful Algal Blooms in the Gulf of Maine Using Ocean Color and Regional Ocean Models https://sciety.org/articles/activity/10.31223/x5br2v

Predicting Unobserved Driver of Regime Shifts in Social-Ecological Systems with Universal Dynamic Equations

Published in EcoEvo Arxiv, 2025

Ecosystems around the world are anticipated to undergo regime shifts as temperatures rise and other climatic and anthropogenic perturbations erode resilience. Forecasting these nonlinear ecosystem dynamics can help stakeholders prepare for rapid changes. One major challenge is that regime shifts can be difficult to predict when driven by unobserved factors—such as illegal fishing from a fishery or unreported poaching in a game reserve. This paper advances scientific machine learning methods, specifically universal dynamic equations (UDEs), to identify changes in an unobserved bifurcation parameter and predict ecosystem regime shifts.

Recommended citation: Rathore, K. J., Buckner, J. H., Meunier, Z. D., Esquivel, J. A., & Watson, J. R. (2025). Predicting unobserved driver of regime shifts in social-ecological systems with universal dynamic equations. https://ecoevorxiv.org/repository/view/11165/

Generating Part-based Global Explnations via Correspondence

Published in IJCAI 2024 Workshop on Explainable Artificial Intelligence (XAI), 2024

Deep learning models are notoriously opaque. Existing explanation methods often focus on localized visual explanations for individual images. Concept-based explanations, while offering global insights, require extensive annotations, incurring significant labeling cost. We propose an approach that leverages user-defined part labels from a limited set of images and efficiently transfers them to a larger dataset. This enables the generation of global symbolic explanations by aggregating part-based local explanations, ultimately providing human-understandable explanations for model decisions on a large scale.

Recommended citation: Rathore, K., & Tadepalli, P. (2025). Generating Part-Based Global Explanations Via Correspondence. arXiv preprint arXiv:2509.15393. https://drive.google.com/file/d/1_1LySYzjUUow22PV4M9FScRstdFIzk89/view

RA2Vec: Distributed Representation of Protein Sequences with Reduced Alphabet Embeddings

Published in Proceedings of the 11th ACM International Conference on Bioinformatics, Computational Biology and Health Informatics (ACM-BCB 2020), 2020

Protein function identification has become an important task due to the rapid growth of sequenced genomes. This work introduces RA2Vec (Reduced Alphabets to Vectors), a novel approach for protein sequence representation that combines reduced amino acid alphabets with Word2Vec-style embeddings. We map protein sequences to reduced alphabets based on hydropathy and conformational similarity, then apply skip-gram models to create distributed vector representations that capture both biochemical properties and sequential context.

Recommended citation: Wijesekara, R. Y., Lahorkar, A., Rathore, K., & Valadi, J. K. (2020). RA2Vec: Distributed representation of protein sequences with reduced alphabet embeddings. In Proceedings of the 11th ACM International Conference on Bioinformatics, Computational Biology and Health Informatics (pp. 1-1). https://dl.acm.org/doi/10.1145/3388440.3414925

A Simple Method of Solution For Multi-label Feature Selection

Published in IEEE International Conference on Electrical, Computer and Communication Technologies, 2019

The study proposes a two-step algorithm for multi-label classification (MLC). It first decomposes the output label space into lower dimensions, then applies feature selection in the reduced space. This approach efficiently handles high-dimensional datasets and reduces computational load.

Recommended citation: Valadi, Jayaraman K., Prasad T. Ovhal, and Kunal J. Rathore. "A simple method of solution for multi-label feature selection." 2019 IEEE International Conference on Electrical, Computer and Communication Technologies (ICECCT). IEEE, 2019. https://ieeexplore.ieee.org/document/8869493