CoMemNet: Contrastive Sampling with Memory Replay Network for Continual Traffic Prediction

2026-05-07

AI Summary

This work addresses the limitations of existing traffic forecasting methods that rely on static graph structures and struggle to adapt to the dynamic evolution of streaming traffic networks, often suffering from catastrophic forgetting. To overcome this, the authors propose CoMemNet, a dual-branch continual learning framework: the online branch performs primary prediction, while the target branch employs a Wasserstein distance–driven dynamic contrastive sampling mechanism to accurately identify critical changing nodes. Coupled with a node-adaptive temporal memory replay buffer (TMRB-N), the framework efficiently consolidates historical knowledge without incurring memory explosion, thereby substantially mitigating forgetting. Evaluated on three large-scale real-world traffic datasets, CoMemNet achieves state-of-the-art performance and introduces two newly curated open-sourced datasets.

Abstract

In recent years, the integration of non-topological space modeling with temporal learning methods has emerged as an effective approach for capturing spatio-temporal information in non-Euclidean graphs. However, most existing methods rely on static underlying graph structures, which are inadequate for capturing the continuously expanding and evolving patterns in streaming traffic networks. To address this challenge, we propose a simple yet efficient dual-branch continual learning framework for traffic prediction, named CoMemNet. The fast-converging Online branch undertakes the primary prediction tasks, while the momentum-updated Target branch extracts historical information using Wasserstein Distance features to create a Dynamic Contrastive Sampler (DC Sampler). This sampler selects a node set with significant dynamic network feature changes for training, effectively mitigating the issue of catastrophic forgetting. Additionally, the backbone incorporates a lightweight Node-Adaptive Temporal Memory Buffer (TMRB-N) to consolidate old knowledge through memory replay and address the risk of memory explosion. Finally, we provide two newly curated open-source datasets. Experimental results demonstrate that CoMemNet achieves state-of-the-art (SOTA) performance across all three large-scale real-world datasets. The code is available at: https://github.com/meiwu5/CoMemNet.

Problem

• continual traffic prediction
• dynamic graph
• catastrophic forgetting
• spatio-temporal learning
• non-Euclidean graphs

Innovation

• Continual Learning
• Dynamic Contrastive Sampling
• Memory Replay
• Traffic Prediction
• Non-Euclidean Graph