comemnet

---
name: comemnet
description: "CoMemNet: Contrastive sampling with Memory Replay Network for continual traffic prediction. Dual-branch architecture with contrastive positive/negative sample generation for efficient memory replay in spatio-temporal forecasting. Use when: traffic prediction, spatio-temporal forecasting, continual learning with memory replay, contrastive CL."
---

# CoMemNet: Contrastive Memory Replay for Traffic Prediction

## Architecture

**Dual-Branch Design**:
- Contrastive positive sample generation
- Contrastive negative sample generation
- Memory replay network for preserving past traffic patterns

## Key Innovation

Contrastive sampling strategy creates informative memory buffers that capture spatio-temporal dependencies more efficiently than random replay.

## Results

Outperforms baselines on multiple traffic prediction benchmarks under domain-incremental settings.

## Paper

- Wu et al., arXiv:2605.05738, 2026