Research
Research Themes
Our transdisciplinary research program is at the intersection of artificial intelligence, data science, and scientific and cultural applications. Our research develops robust and uncertainty-aware machine learning methods and applies them to challenges in Earth and space sciences, climate, mineral exploration, language, culture, and human-centred AI.
A central focus is the development of deep-learning and Bayesian frameworks for uncertainty-aware decision-making, complemented by research on foundation models, multimodal LLMs, and responsible AI.
Our research is organised around three interconnected themes:
Research themes spanning ML methodologies, Earth and space sciences, and digital humanities and human-centred AI.
1. AI-ML Methodologies
We develop machine-learning methods for robust prediction, representation learning, and decision-making, with particular emphasis on Bayesian deep learning, neuroevolution, ensemble learning, data augmentation, and uncertainty quantification.
Our early work developed neuroevolutionary methods for dynamic time-series forecasting and modular pattern recognition (Chandra et al., 2017; Chandra et al., 2018; Chandra and Cripps, 2018). We subsequently developed GAN-based data augmentation for limited and imbalanced datasets (Sharma et al., 2022) and methods for extreme-event forecasting (Hua et al., 2025).
A major research direction is Bayesian deep learning, including scalable Bayesian neural networks using parallel-tempering MCMC and parallel computing (Chandra et al., 2019), subsequently extended to Bayesian graph neural networks, autoencoders, and transfer learning (Chandra et al., 2021; Chandra et al., 2022; Chandra and Kapoor, 2020).
Current work includes uncertainty-aware data imputation, spatiotemporal modelling, and the development of BayesClustering for uncertainty-aware clustering and representation learning. We are also exploring the integration of Bayesian inference with multimodal and foundation models.
2. Earth, Climate and Space
We apply machine learning, Bayesian inference, and remote sensing to environmental modelling, mineral exploration, climate extremes, and planetary science.
Our work has developed machine-learning approaches for geological lineament detection and mineral exploration from satellite imagery (Farahbakhsh et al., 2020) and deep-learning approaches for lithological mapping (Shirmard et al., 2022). Current research combines satellite imagery, geochemical observations, and spatial machine learning for critical-mineral exploration, together with open-source land-cover mapping tools for Fiji (Gurjar et al., 2026).
We have also developed machine-learning methods for climate extremes, tropical cyclone forecasting, floods, and streamflow modelling (Chandra and Dayal, 2015; Deo and Chandra, 2019; Sutar and Chandra, 2025; Kapoor and Chandra, 2026).
We also developed the Bayeslands and Bayesreef frameworks for Bayesian inference and uncertainty quantification in geoscientific models of landscape and reef evolution (Chandra et al., 2019; Chandra et al., 2020).
More recently, we have extended this research to planetary exploration, developing deep-learning methods for crater detection on the Moon and Mars (Ma et al., 2026) and unsupervised learning for mineral mapping from Martian imagery (Lovelock and Chandra, 2025).
Current research investigates multimodal AI for planetary science, combining orbital imagery, terrain models, geological information, and scientific literature to analyse Martian environments, including dust storms, geology, mineral resources, and candidate landing sites.
Our long-term goal is to develop trustworthy AI systems that support autonomous scientific discovery and future robotic and human exploration of the Moon and Mars.
3. Digital Humanities and Human-Centred AI
We explore how large language models and foundation models can support the study of language, culture, philosophy, religion, cinema, and music, while developing human-centred approaches to the responsible use of AI.
Our research has developed frameworks for evaluating machine translation across Sanskrit, Mandarin, and Indian languages, comparing human translation with Google Translate and large language models (Shukla et al., 2023; Wang et al., 2025; Chandra et al., 2025).
We pioneered computational approaches to ancient religious and philosophical texts, including the Bhagavad Gita and Upanishads (Chandra and Kulkarni, 2022; Chandra and Ranjan, 2022), and developed methods for metaphor detection across religious texts (Chandra et al., 2024).
Current work includes Vedanta-RAG, retrieval-augmented generation for Hindu philosophical knowledge, together with research on hallucination, retrieval, and factual grounding in language models.
We also apply language models to cinema and music, including longitudinal analysis of violence and narrative patterns in Hollywood movies (Chandra and Ren, 2025).
A related focus is human-centred and responsible AI, examining how language models represent and reason about concepts such as fairness, personality, consciousness, values, and ethical decision-making. Current research explores philosophical concepts including Ahimsa and Nyaya as potential foundations for ethical decision-making in agentic AI.
Together, this research develops computational approaches to human culture while investigating how AI systems can remain grounded in human values, cultural knowledge, and responsible decision-making.
Research Vision
Our goal is to develop trustworthy, uncertainty-aware, and human-centred AI systems that combine data, scientific knowledge, and human perspectives.
By bringing together advances in Bayesian AI, deep learning, foundation models, remote sensing, and computational humanities, we aim to develop AI that can learn from diverse sources, quantify uncertainty, and support reliable human decision-making across scientific, environmental, cultural, and societal domains.