Where to Measure: Epistemic Uncertainty-Based Sensor Placement with ConvCNPs

TL;DR

This paper introduces a novel sensor placement method using epistemic uncertainty estimates from ConvCNPs, improving data collection efficiency for spatio-temporal systems by focusing on model uncertainty rather than total uncertainty.

Contribution

It extends ConvCNPs with MDN outputs to estimate epistemic uncertainty and proposes an acquisition function based on this for better sensor placement.

Findings

Epistemic uncertainty-based placement reduces model error more effectively.

Extending ConvCNPs with MDNs enables explicit epistemic uncertainty estimation.

Preliminary results show improved sensor placement over total uncertainty methods.

Abstract

Accurate sensor placement is critical for modeling spatio-temporal systems such as environmental and climate processes. Neural Processes (NPs), particularly Convolutional Conditional Neural Processes (ConvCNPs), provide scalable probabilistic models with uncertainty estimates, making them well-suited for data-driven sensor placement. However, existing approaches rely on total predictive uncertainty, which conflates epistemic and aleatoric components, that may lead to suboptimal sensor selection in ambiguous regions. To address this, we propose expected reduction in epistemic uncertainty as a new acquisition function for sensor placement. To enable this, we extend ConvCNPs with a Mixture Density Networks (MDNs) output head for epistemic uncertainty estimation. Preliminary results suggest that epistemic uncertainty driven sensor placement more effectively reduces model error than approaches based on overall uncertainty.