Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Spatial-Temporal Feedback Diffusion Guidance for Controlled Traffic Imputation

Xiaowei Mao, Huihu Ding, Yan Lin, Tingrui Wu, Shengnan Guo, Dazhuo Qiu, Feiling Fang, Jilin Hu, Huaiyu Wan

TL;DR

This paper tackles missing values in spatial-temporal traffic data by introducing FENCE, a diffusion-based imputation framework that dynamically adjusts the conditioning guidance during generation. The key innovation is a posterior-driven, adaptive guidance scale that increases when generated values diverge from observations and decreases when alignment is strong, implemented globally and at cluster level to exploit spatial-temporal correlations. FENCE employs a two-stage training regime (unconditional prior learning followed by conditional fine-tuning) and a cluster-aware strategy that groups nodes via dynamic attention scores to stabilize guidance. Experiments on PEMS04/07/08 show FENCE achieving state-of-the-art imputation accuracy under challenging missing patterns, demonstrating practical impact for improving data quality in intelligent transportation systems.

Abstract

Imputing missing values in spatial-temporal traffic data is essential for intelligent transportation systems. Among advanced imputation methods, score-based diffusion models have demonstrated competitive performance. These models generate data by reversing a noising process, using observed values as conditional guidance. However, existing diffusion models typically apply a uniform guidance scale across both spatial and temporal dimensions, which is inadequate for nodes with high missing data rates. Sparse observations provide insufficient conditional guidance, causing the generative process to drift toward the learned prior distribution rather than closely following the conditional observations, resulting in suboptimal imputation performance. To address this, we propose FENCE, a spatial-temporal feedback diffusion guidance method designed to adaptively control guidance scales during imputation. First, FENCE introduces a dynamic feedback mechanism that adjusts the guidance scale based on the posterior likelihood approximations. The guidance scale is increased when generated values diverge from observations and reduced when alignment improves, preventing overcorrection. Second, because alignment to observations varies across nodes and denoising steps, a global guidance scale for all nodes is suboptimal. FENCE computes guidance scales at the cluster level by grouping nodes based on their attention scores, leveraging spatial-temporal correlations to provide more accurate guidance. Experimental results on real-world traffic datasets show that FENCE significantly enhances imputation accuracy.

Spatial-Temporal Feedback Diffusion Guidance for Controlled Traffic Imputation

TL;DR

This paper tackles missing values in spatial-temporal traffic data by introducing FENCE, a diffusion-based imputation framework that dynamically adjusts the conditioning guidance during generation. The key innovation is a posterior-driven, adaptive guidance scale that increases when generated values diverge from observations and decreases when alignment is strong, implemented globally and at cluster level to exploit spatial-temporal correlations. FENCE employs a two-stage training regime (unconditional prior learning followed by conditional fine-tuning) and a cluster-aware strategy that groups nodes via dynamic attention scores to stabilize guidance. Experiments on PEMS04/07/08 show FENCE achieving state-of-the-art imputation accuracy under challenging missing patterns, demonstrating practical impact for improving data quality in intelligent transportation systems.

Abstract

Imputing missing values in spatial-temporal traffic data is essential for intelligent transportation systems. Among advanced imputation methods, score-based diffusion models have demonstrated competitive performance. These models generate data by reversing a noising process, using observed values as conditional guidance. However, existing diffusion models typically apply a uniform guidance scale across both spatial and temporal dimensions, which is inadequate for nodes with high missing data rates. Sparse observations provide insufficient conditional guidance, causing the generative process to drift toward the learned prior distribution rather than closely following the conditional observations, resulting in suboptimal imputation performance. To address this, we propose FENCE, a spatial-temporal feedback diffusion guidance method designed to adaptively control guidance scales during imputation. First, FENCE introduces a dynamic feedback mechanism that adjusts the guidance scale based on the posterior likelihood approximations. The guidance scale is increased when generated values diverge from observations and reduced when alignment improves, preventing overcorrection. Second, because alignment to observations varies across nodes and denoising steps, a global guidance scale for all nodes is suboptimal. FENCE computes guidance scales at the cluster level by grouping nodes based on their attention scores, leveraging spatial-temporal correlations to provide more accurate guidance. Experimental results on real-world traffic datasets show that FENCE significantly enhances imputation accuracy.
Paper Structure (32 sections, 49 equations, 5 figures, 3 tables, 1 algorithm)

This paper contains 32 sections, 49 equations, 5 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. Methods
  5. Guidance in Diffusion Models for Traffic Imputation
  6. Spatial-Temporal Feedback Diffusion Guidance
  7. Posterior-Driven Dynamic Guidance Scaling
  8. Posterior Likelihood Estimation
  9. Cluster-Aware Feedback Guidance
  10. Training and Inference
  11. Training.
  12. Inference.
  13. Experiments
  14. Experimental Settings
  15. Dateset.
  16. ...and 17 more sections

Figures (5)

  • Figure 1: Motivation for FENCE. Unlike CSDI, which uses a fixed guidance scale, FENCE dynamically adjusts guidance scale based on consistency with observed data.
  • Figure 2: FENCE performs imputation by estimating both conditional and unconditional scores. It dynamically adjusts the guidance scale at each step by evaluating the posterior likelihood, controlling the scales of the conditional guidance strength to ensure consistency with observed data.
  • Figure 3: Ablation study.
  • Figure 4: Effect of hyperparameters.
  • Figure 5: Case Study