Nisha M. Pillai

Assistant Research Professor | Machine Learning, Computer Vision, NLP

Mississippi State University · Bagley College of Engineering

My research addresses core problems in machine learning, computer vision, spatio-temporal modeling, and natural language processing. A common thread across all of my work is building AI systems that learn reliably from limited or noisy data, produce outputs that domain experts can understand and act on, and remain robust when deployed outside controlled settings. I validate my methods on high-impact problems in disease surveillance, environmental monitoring, health, and language understanding. Current and pending federal funding includes proposals to USDA NIFA, NSF, and NIH spanning machine learning, trustworthy AI, and multimodal health systems.

Learning Under Limited Data.

A core challenge in deploying machine learning to real-world settings is the scarcity of labeled training data. In collaboration with the USDA Agricultural Research Service, my group addresses this through policy-driven transfer learning and diffusion-based data augmentation, validated on UAV-based frameworks using RGB, thermal, and multispectral imaging for early detection of disease and health indicators in livestock. Both methods were published at peer-reviewed IEEE venues in 2025.

Natural Language Processing and Grounded Language Learning.
My doctoral research addressed grounded language acquisition, building systems that learn to associate natural language with perceptual representations from limited annotated data. This work produced publications at AAAI 2018 and IEEE BigData 2020, among other venues, and introduced active learning and variational autoencoder approaches that significantly reduced annotation requirements. I continue to work in NLP, with current students applying large language models to intrusion detection and cybersecurity, and developing speech-based AI systems for dementia detection with a focus on cross-cultural generalization and equitable diagnostic performance across linguistic populations.

Explainable AI and Generative Modeling.

In many applied settings, a prediction that cannot be explained is not actionable. My work in this area develops AI methods that are both accurate and interpretable. I developed a Bayesian optimization framework that encodes high-dimensional biological datasets into a low-dimensional latent space using autoencoders, then searches that space to identify synthetic data variants with desired properties, published at IEEE EMBC 2024. I have also applied ensemble learning and sensitivity analysis to identify interpretable combinations of features that drive outcomes in complex biological and environmental datasets, with results published in Animal Microbiome. My open-source EndToEndML pipeline makes these methods accessible to researchers without programming expertise.

Spatio-Temporal Modeling and Graph Neural Networks.
Many real-world processes unfold across both geographic space and time, and modeling them requires architectures that capture both dimensions simultaneously. My work in this area focuses on disease surveillance, where I develop interpretable spatio-temporal models that identify the ecological and environmental drivers of outbreak spread. A recent collaboration produced a symbolic regression model for Vesicular Stomatitis Virus transmission across the United States and Mexico, accepted at ACM SIGSPATIAL 2025. I am currently extending this to spatio-temporal graph neural networks for both VSV and Highly Pathogenic Avian Influenza, and developing early warning systems that integrate remote sensing, climate, and livestock movement data.

Across all of this work, my guiding principle is that the value of an AI system is measured not by model performance alone, but by whether a domain expert can understand and act on what it tells them.

Research Interests

Machine Learning
Computer Vision
Natural Language Processing
Explainable AI
Spatio-Temporal Modeling
Human-Robot Interaction
Generative Modeling
Robotics
AI for Health

Research Experience

Assistant Research Professor — Machine Learning, Computer Vision, NLP, Spatio-Temporal Modeling

Mississippi State University  ·  April 2024 – Present

  • Guiding collaborative investigation into neuroimaging and language processing biomarkers for brain aging.
  • Designing computer vision architectures for animal monitoring in resource-constrained environments.
  • Collaborating with Civil Engineering on ML methods for identifying job satisfaction factors.
  • PI on research proposals to USDA and NSF across agriculture, NLP, wireless systems, and security.
  • Mentoring graduate and undergraduate students across security, software engineering, CV, and NLP.

Research Scientist I — Deep Learning, Machine Learning

Mississippi State University  ·  Feb 2023 – April 2024

  • Developed data-driven frameworks for identifying microbiome signatures predictive of reduced multidrug resistance.
  • Built open-source end-to-end ML pipeline for non-programmers to analyze complex biological data.

Post-doctoral Associate — Graph Neural Networks, Deep Learning

Mississippi State University  ·  Sep 2021 – Jan 2023

  • Developed GNN-based systems to analyze factors affecting infection rates in pasture-raised poultry.
  • Designed ensemble framework to predict physicochemical soil/fecal parameters influencing pathogen prevalence.
  • Collaborated on Cyber-Physical security surveillance framework using digital twin anomaly detection.

Research Assistant — ML, NLP, Robotics

UMBC  ·  May 2016 – Aug 2021

  • Designed language learning systems jointly learning linguistic concepts and visual percepts via probabilistic algorithms.
  • Proposed unsupervised GMM-based active learning approach reducing annotation data by 5% across complex settings.
  • Developed category-free unsupervised deep generative VAE with minimum grounding score of 0.45 (baseline <0.1).