Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

TL;DR

This work proposes a unified geospatial discovery framework that integrates active learning, online meta-learning, and concept-guided reasoning, and introduces two key innovations built on a shared notion of *concept relevance*, which captures how domain-specific factors influence target presence.

Abstract

In many real-world settings, such as environmental monitoring, disaster response, or public health, with costly and difficult data collection and dynamic environments, strategically sampling from unobserved regions is essential for efficiently uncovering hidden targets under tight resource constraints. Yet, sparse and biased geospatial ground truth limits the applicability of existing learning-based methods, such as reinforcement learning. To address this, we propose a unified geospatial discovery framework that integrates active learning, online meta-learning, and concept-guided reasoning. Our approach introduces two key innovations built on a shared notion of *concept relevance*, which captures how domain-specific factors influence target presence: a *concept-weighted uncertainty sampling strategy*, where uncertainty is modulated by learned relevance based on readily-available domain-specific concepts (e.g., land cover, source proximity); and a *relevance-aware meta-batch formation strategy* that promotes semantic diversity during online-meta updates, improving generalization in dynamic environments. Our experiments include testing on a real-world dataset of cancer-causing PFAS (Per- and polyfluoroalkyl substances) contamination, showcasing our method's reliability at uncovering targets with limited data and a varying environment.

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

TL;DR

This work proposes a unified geospatial discovery framework that integrates active learning, online meta-learning, and concept-guided reasoning, and introduces two key innovations built on a shared notion of *concept relevance*, which captures how domain-specific factors influence target presence.

Abstract

In many real-world settings, such as environmental monitoring, disaster response, or public health, with costly and difficult data collection and dynamic environments, strategically sampling from unobserved regions is essential for efficiently uncovering hidden targets under tight resource constraints. Yet, sparse and biased geospatial ground truth limits the applicability of existing learning-based methods, such as reinforcement learning. To address this, we propose a unified geospatial discovery framework that integrates active learning, online meta-learning, and concept-guided reasoning. Our approach introduces two key innovations built on a shared notion of *concept relevance*, which captures how domain-specific factors influence target presence: a *concept-weighted uncertainty sampling strategy*, where uncertainty is modulated by learned relevance based on readily-available domain-specific concepts (e.g., land cover, source proximity); and a *relevance-aware meta-batch formation strategy* that promotes semantic diversity during online-meta updates, improving generalization in dynamic environments. Our experiments include testing on a real-world dataset of cancer-causing PFAS (Per- and polyfluoroalkyl substances) contamination, showcasing our method's reliability at uncovering targets with limited data and a varying environment.
Paper Structure (59 sections, 5 theorems, 43 equations, 13 figures, 13 tables, 3 algorithms)

This paper contains 59 sections, 5 theorems, 43 equations, 13 figures, 13 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Problem Formulation
  3. Related Work
  4. RL-based Approaches for Active Geospatial Search (AGS)
  5. Active, Online, Meta, and Lifelong Learning
  6. Methodology
  7. Overview
  8. Concept Encoder
  9. Relevance Encoder and Decoder
  10. Meta-training Set Formation
  11. Online-Meta Policy Update Rule at t-th Query Step
  12. Sampling Strategy During Training
  13. Sampling Strategy During Inference
  14. Experiments
  15. Evaluation Metrics
  16. ...and 44 more sections

Key Result

Proposition 4.1

Optimizing eq:obj is equivalent to minimizing the following objective:

Figures (13)

  • Figure 1: Illustration of geospatial learning under sparse and costly ground truth, where each query depletes a limited sampling budget and an adaptive policy must prioritize target discovery. The goal is to optimize data collection for continual model learning and maximal target discovery.
  • Figure 2: Block Diagram of Framework
  • Figure 3: Meta-training Set Formation
  • Figure 4: Active Sampling Strategy.
  • Figure 5: Performance and exploration behavior.
  • ...and 8 more figures

Theorems & Definitions (8)

  • Proposition 4.1
  • Theorem 4.2
  • Theorem 4.3
  • proof
  • Theorem 8.1
  • proof
  • Theorem 9.1
  • proof