Adapting Actively on the Fly: Relevance-Guided Online Meta-Learning with Latent Concepts for Geospatial Discovery

Jowaria Khan, Anindya Sarkar, Yevgeniy Vorobeychik, Elizabeth Bondi-Kelly

Keywords: Geospatial Modeling, Active Learning

Abstract:

In many real-world settings, such as environmental monitoring, disaster response, or public health, where data is costly and difficult to collect, strategically sampling from unobserved regions is essential for uncovering hidden risks or targets under tight resource constraints. Moreover, real-world geospatial data is often sparse, noisy, and geographically skewed, rendering most existing learning-based methods
impractical. To address this, we propose a unified geospatial discovery framework that integrates active learning, online meta-learning, and concept-guided reasoning to support efficient, real-time decision-making under extreme data scarcity. Our approach introduces two key innovations: a relevance-uncertainty guided sampling strategy that uses structured relevance vectors based on domain-specific concepts (e.g., spectral channels, industrial proximity), enabling interpretable and adaptive sample selection; and a relevance-aware meta-batch formation strategy that promotes semantic diversity during online-meta updates, improving generalization in dynamic environments. Our experiments include actively searching for a specific land cover type under sparse training conditions and a strict sampling budget, as well as identifying cancer-causing PFAS (Per- and polyfluoroalkyl substances) contamination hotspots in U.S. surface water bodies, a critical, real-world public health and environmental problem. Despite limited observations and significant landscape shifts, our method reliably uncovers target land covers and contamination zones and adapts across space and time, showcasing its scalability, robustness, and potential to accelerate discovery in data-limited, high-stakes environments.

Primary Area: transfer learning, meta learning, and lifelong learning

Submission Number: 10184