Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SimFair: Physics-Guided Fairness-Aware Learning with Simulation Models

Zhihao Wang, Yiqun Xie, Zhili Li, Xiaowei Jia, Zhe Jiang, Aolin Jia, Shuo Xu

TL;DR

SimFair addresses location-based fairness under regional distribution shifts by leveraging physics-based mechanistic simulations to guide learning in the absence of labels in new regions. It introduces inverse modeling via an invertible network to couple simulation outputs with data-driven predictors, adds a preliminary test-region fairness measure, and enforces dual-fairness consistency together with physics-based losses. Across CMEM and MODTRAN-based temperature prediction tasks on AT1, AT2, and LST datasets, SimFair improves fairness while maintaining or enhancing overall predictive accuracy, demonstrating robustness to diverse partitions. The work provides a general, physics-guided framework for fair learning in remote sensing and spatiotemporal prediction, with potential applicability to broader domains that employ simulation models.

Abstract

Fairness-awareness has emerged as an essential building block for the responsible use of artificial intelligence in real applications. In many cases, inequity in performance is due to the change in distribution over different regions. While techniques have been developed to improve the transferability of fairness, a solution to the problem is not always feasible with no samples from the new regions, which is a bottleneck for pure data-driven attempts. Fortunately, physics-based mechanistic models have been studied for many problems with major social impacts. We propose SimFair, a physics-guided fairness-aware learning framework, which bridges the data limitation by integrating physical-rule-based simulation and inverse modeling into the training design. Using temperature prediction as an example, we demonstrate the effectiveness of the proposed SimFair in fairness preservation.

SimFair: Physics-Guided Fairness-Aware Learning with Simulation Models

TL;DR

SimFair addresses location-based fairness under regional distribution shifts by leveraging physics-based mechanistic simulations to guide learning in the absence of labels in new regions. It introduces inverse modeling via an invertible network to couple simulation outputs with data-driven predictors, adds a preliminary test-region fairness measure, and enforces dual-fairness consistency together with physics-based losses. Across CMEM and MODTRAN-based temperature prediction tasks on AT1, AT2, and LST datasets, SimFair improves fairness while maintaining or enhancing overall predictive accuracy, demonstrating robustness to diverse partitions. The work provides a general, physics-guided framework for fair learning in remote sensing and spatiotemporal prediction, with potential applicability to broader domains that employ simulation models.

Abstract

Fairness-awareness has emerged as an essential building block for the responsible use of artificial intelligence in real applications. In many cases, inequity in performance is due to the change in distribution over different regions. While techniques have been developed to improve the transferability of fairness, a solution to the problem is not always feasible with no samples from the new regions, which is a bottleneck for pure data-driven attempts. Fortunately, physics-based mechanistic models have been studied for many problems with major social impacts. We propose SimFair, a physics-guided fairness-aware learning framework, which bridges the data limitation by integrating physical-rule-based simulation and inverse modeling into the training design. Using temperature prediction as an example, we demonstrate the effectiveness of the proposed SimFair in fairness preservation.
Paper Structure (29 sections, 10 equations, 8 figures, 5 tables)

This paper contains 29 sections, 10 equations, 8 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Problem Definition
  3. Formulation of location-based fair learning.
  4. Method
  5. Physics-based Mechanistic Models
  6. Physics-Model 1 (PM1):
  7. Physics-Model 2 (PM2):
  8. SimFair: Simulation-Enabled Fair Learning
  9. Inverse Modeling for Learning.
  10. Preliminary Fairness on Test Region.
  11. Dual-Fairness Consistency.
  12. Improvements with Physics-guided Predictions.
  13. Deep Networks
  14. Experiments
  15. In-Situ and Remote Sensing Datasets
  16. ...and 14 more sections

Figures (8)

  • Figure 1: An illustration of simulation-model-guided fairness-aware learning.
  • Figure 2: Illustrative example of dual-fairness consistency.
  • Figure 3: Spatial distributions of training and testing data.
  • Figure 4: Example approximation results on CMEM.
  • Figure 5: Distributions of absolute errors in AT2 & LST
  • ...and 3 more figures

Theorems & Definitions (3)

  • Definition 1: Spatiotemporal (ST) domain
  • Definition 2: Location-based fairness measure
  • Definition 3: Dual-fairness consistency