Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FairSTG: Countering performance heterogeneity via collaborative sample-level optimization

Gengyu Lin, Zhengyang Zhou, Qihe Huang, Kuo Yang, Shifen Cheng, Yang Wang

TL;DR

This work addresses the problem of performance fairness in spatiotemporal forecasting for urban environments, where sample-level heterogeneity leads to underrepresented regions receiving poorer predictions. It introduces FairSTG, a model-agnostic framework that combines a fairness recognizer with collaborative representation enhancement to transfer knowledge from well-learned samples to challenging ones, guided by a variance-based fairness objective and a self-supervised fairness signal. Empirical results on four real-world datasets show that FairSTG significantly improves sample-level fairness (lower MAE-var and MAPE-var) while maintaining competitive forecasting accuracy, with ablations and case studies illustrating effectiveness in both spatial and temporal heterogeneity. The proposed approach offers a scalable, transferable paradigm for fair urban computing and resource allocation, with potential applicability to a broad range of mobile computing tasks beyond traffic and air quality forecasting.

Abstract

Spatiotemporal learning plays a crucial role in mobile computing techniques to empower smart cites. While existing research has made great efforts to achieve accurate predictions on the overall dataset, they still neglect the significant performance heterogeneity across samples. In this work, we designate the performance heterogeneity as the reason for unfair spatiotemporal learning, which not only degrades the practical functions of models, but also brings serious potential risks to real-world urban applications. To fix this gap, we propose a model-independent Fairness-aware framework for SpatioTemporal Graph learning (FairSTG), which inherits the idea of exploiting advantages of well-learned samples to challenging ones with collaborative mix-up. Specifically, FairSTG consists of a spatiotemporal feature extractor for model initialization, a collaborative representation enhancement for knowledge transfer between well-learned samples and challenging ones, and fairness objectives for immediately suppressing sample-level performance heterogeneity. Experiments on four spatiotemporal datasets demonstrate that our FairSTG significantly improves the fairness quality while maintaining comparable forecasting accuracy. Case studies show FairSTG can counter both spatial and temporal performance heterogeneity by our sample-level retrieval and compensation, and our work can potentially alleviate the risks on spatiotemporal resource allocation for underrepresented urban regions.

FairSTG: Countering performance heterogeneity via collaborative sample-level optimization

TL;DR

This work addresses the problem of performance fairness in spatiotemporal forecasting for urban environments, where sample-level heterogeneity leads to underrepresented regions receiving poorer predictions. It introduces FairSTG, a model-agnostic framework that combines a fairness recognizer with collaborative representation enhancement to transfer knowledge from well-learned samples to challenging ones, guided by a variance-based fairness objective and a self-supervised fairness signal. Empirical results on four real-world datasets show that FairSTG significantly improves sample-level fairness (lower MAE-var and MAPE-var) while maintaining competitive forecasting accuracy, with ablations and case studies illustrating effectiveness in both spatial and temporal heterogeneity. The proposed approach offers a scalable, transferable paradigm for fair urban computing and resource allocation, with potential applicability to a broad range of mobile computing tasks beyond traffic and air quality forecasting.

Abstract

Spatiotemporal learning plays a crucial role in mobile computing techniques to empower smart cites. While existing research has made great efforts to achieve accurate predictions on the overall dataset, they still neglect the significant performance heterogeneity across samples. In this work, we designate the performance heterogeneity as the reason for unfair spatiotemporal learning, which not only degrades the practical functions of models, but also brings serious potential risks to real-world urban applications. To fix this gap, we propose a model-independent Fairness-aware framework for SpatioTemporal Graph learning (FairSTG), which inherits the idea of exploiting advantages of well-learned samples to challenging ones with collaborative mix-up. Specifically, FairSTG consists of a spatiotemporal feature extractor for model initialization, a collaborative representation enhancement for knowledge transfer between well-learned samples and challenging ones, and fairness objectives for immediately suppressing sample-level performance heterogeneity. Experiments on four spatiotemporal datasets demonstrate that our FairSTG significantly improves the fairness quality while maintaining comparable forecasting accuracy. Case studies show FairSTG can counter both spatial and temporal performance heterogeneity by our sample-level retrieval and compensation, and our work can potentially alleviate the risks on spatiotemporal resource allocation for underrepresented urban regions.
Paper Structure (27 sections, 18 equations, 7 figures, 6 tables)

This paper contains 27 sections, 18 equations, 7 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Spatiotemporal learning
  4. Fairness-aware machine learning
  5. Problem Formulation
  6. Methodology
  7. Spatiotemporal feature extractor
  8. Fairness-aware spatiotemporal architecture
  9. Fairness recognizer
  10. Collaborative feature enhancement
  11. Fairness-aware learning objective
  12. Output module
  13. Optimization objective
  14. Training strategy
  15. Experiments
  16. ...and 12 more sections

Figures (7)

  • Figure 1: The unfairness across spatial and temporal ranges. (a) The mean MAPE of each sensor in PEMS-BAY dataset generated by MTGNN, with red boxes indicating regions with significant errors. (b) The curve shows the traffic flow data and the corresponding predictions from MTGNN on sensor #64 in PEMS-BAY dataset. It can be observed that the prediction performance varies significantly at different time stamps even for the same sensor.
  • Figure 2: Framework overview of FairSTG. The spatiotemporal feature extractor learns spatiotemporal representations from the original ST Graph. The fairness recognizer mines the learning difficulty and generates fairness signals in a self-supervised manner, and the collaborative feature enhancement adaptively transfers advantageous features from easy set to challenging set. Finally, the output module transforms the fused representations and produces the predictions.
  • Figure 3: Forecasting and fairness performance comparison at each horizon. The top and bottom lines respectively indicate the prediction accuracy and fairness performances.
  • Figure 4: The accuracy of fairness recognizer in self-supervised classification.
  • Figure 5: Case study in compensatory samples. (a) The raw time series of a challenging sample and its compensatory samples. (b) The sampling time stamps and nodes for these samples.
  • ...and 2 more figures

Theorems & Definitions (3)

  • Definition 1: Spatiotemporal graph
  • Definition 2: Fairness metrics in spatiotemporal forecasting
  • Definition 3: Easy samples and challenging samples