Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Online Feedback Efficient Active Target Discovery in Partially Observable Environments

Anindya Sarkar, Binglin Ji, Yevgeniy Vorobeychik

TL;DR

This work tackles the high cost of sampling in partially observable environments for targeted discovery. It introduces DiffATD, a training-free diffusion-guided approach that builds a posterior belief over unobserved regions via Tweedie denoising and diffusion scores, combining maximum-entropy exploration with an online reward-driven exploitation under a fixed budget. DiffATD demonstrates strong, transferable performance across medical imaging, remote sensing, and ecological domains, outperforming classical baselines and approaching fully supervised or fully observable benchmarks without task-specific pretraining. The method offers interpretability and a principled Bayesian foundation, with robust performance under noise and scalable computation across moderate-to-large search spaces.

Abstract

In various scientific and engineering domains, where data acquisition is costly--such as in medical imaging, environmental monitoring, or remote sensing--strategic sampling from unobserved regions, guided by prior observations, is essential to maximize target discovery within a limited sampling budget. In this work, we introduce Diffusion-guided Active Target Discovery (DiffATD), a novel method that leverages diffusion dynamics for active target discovery. DiffATD maintains a belief distribution over each unobserved state in the environment, using this distribution to dynamically balance exploration-exploitation. Exploration reduces uncertainty by sampling regions with the highest expected entropy, while exploitation targets areas with the highest likelihood of discovering the target, indicated by the belief distribution and an incrementally trained reward model designed to learn the characteristics of the target. DiffATD enables efficient target discovery in a partially observable environment within a fixed sampling budget, all without relying on any prior supervised training. Furthermore, DiffATD offers interpretability, unlike existing black--box policies that require extensive supervised training. Through extensive experiments and ablation studies across diverse domains, including medical imaging, species discovery, and remote sensing, we show that DiffATD performs significantly better than baselines and competitively with supervised methods that operate under full environmental observability.

Online Feedback Efficient Active Target Discovery in Partially Observable Environments

TL;DR

This work tackles the high cost of sampling in partially observable environments for targeted discovery. It introduces DiffATD, a training-free diffusion-guided approach that builds a posterior belief over unobserved regions via Tweedie denoising and diffusion scores, combining maximum-entropy exploration with an online reward-driven exploitation under a fixed budget. DiffATD demonstrates strong, transferable performance across medical imaging, remote sensing, and ecological domains, outperforming classical baselines and approaching fully supervised or fully observable benchmarks without task-specific pretraining. The method offers interpretability and a principled Bayesian foundation, with robust performance under noise and scalable computation across moderate-to-large search spaces.

Abstract

In various scientific and engineering domains, where data acquisition is costly--such as in medical imaging, environmental monitoring, or remote sensing--strategic sampling from unobserved regions, guided by prior observations, is essential to maximize target discovery within a limited sampling budget. In this work, we introduce Diffusion-guided Active Target Discovery (DiffATD), a novel method that leverages diffusion dynamics for active target discovery. DiffATD maintains a belief distribution over each unobserved state in the environment, using this distribution to dynamically balance exploration-exploitation. Exploration reduces uncertainty by sampling regions with the highest expected entropy, while exploitation targets areas with the highest likelihood of discovering the target, indicated by the belief distribution and an incrementally trained reward model designed to learn the characteristics of the target. DiffATD enables efficient target discovery in a partially observable environment within a fixed sampling budget, all without relying on any prior supervised training. Furthermore, DiffATD offers interpretability, unlike existing black--box policies that require extensive supervised training. Through extensive experiments and ablation studies across diverse domains, including medical imaging, species discovery, and remote sensing, we show that DiffATD performs significantly better than baselines and competitively with supervised methods that operate under full environmental observability.
Paper Structure (45 sections, 4 theorems, 43 equations, 28 figures, 21 tables, 2 algorithms)

This paper contains 45 sections, 4 theorems, 43 equations, 28 figures, 21 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. Problem Formulation
  5. Solution Approach
  6. Experiments
  7. Active Discovery of Objects from Overhead Imagery
  8. Active Discovery of Species
  9. Active Discovery of Skin Disease
  10. Active Discovery of Bone Suppression
  11. Qualitative Comparison with GA Approach
  12. Effect of $\kappa(\mathcal{B}$)
  13. Comparison with Supervised and Fully Observable Approach
  14. Visualizing DiffATD's Exploration Behavior at Different Stage
  15. Comparison With Vision-Language Model
  16. ...and 30 more sections

Key Result

Proposition 1

Maximizing $I(\hat{x}_t; x \mid Q_t, \tilde{x}_{t-1})$ w.r.t a measurement location $q_t$ is equivalent to maximizing the marginal entropy of the estimated search space $\hat{x}_t$, i.e.,

Figures (28)

  • Figure 1: DiffATD uncovers more target pixels (e.g., plane) while exhibiting higher uncertainty in the search space. In contrast, DiffATD without exploitation explores the environment more effectively, resulting in lower entropy but identifies fewer target pixels.
  • Figure 2: An Overview of DiffATD.
  • Figure 3: Active Discovery of Plane, Playground.
  • Figure 4: ATD of Species (purple $\rightarrow$ targets).
  • Figure 5: Active Discovery of Skin Disease.
  • ...and 23 more figures

Theorems & Definitions (7)

  • Proposition 1
  • Theorem 1
  • Theorem 2
  • Theorem 3
  • proof
  • proof
  • proof