Lightweight and Interpretable Transformer via Mixed Graph Algorithm Unrolling for Traffic Forecast

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Qi, Ji, Do, Tam Thuc, Liu, Mingxiao, Pan, Zhuoshi, Li, Yuzhe, Cheung, Gene, Zhao, H. Vicky
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912869499535360
author Qi, Ji
Do, Tam Thuc
Liu, Mingxiao
Pan, Zhuoshi
Li, Yuzhe
Cheung, Gene
Zhao, H. Vicky
author_facet Qi, Ji
Do, Tam Thuc
Liu, Mingxiao
Pan, Zhuoshi
Li, Yuzhe
Cheung, Gene
Zhao, H. Vicky
contents Unlike conventional "black-box" transformers with classical self-attention mechanism, we build a lightweight and interpretable transformer-like neural net by unrolling a mixed-graph-based optimization algorithm to forecast traffic with spatial and temporal dimensions. We construct two graphs: an undirected graph $\mathcal{G}^u$ capturing spatial correlations across geography, and a directed graph $\mathcal{G}^d$ capturing sequential relationships over time. We predict future samples of signal $\mathbf{x}$, assuming it is "smooth" with respect to both $\mathcal{G}^u$ and $\mathcal{G}^d$, where we design new $\ell_2$ and $\ell_1$-norm variational terms to quantify and promote signal smoothness (low-frequency reconstruction) on a directed graph. We design an iterative algorithm based on alternating direction method of multipliers (ADMM), and unroll it into a feed-forward network for data-driven parameter learning. We periodically insert graph learning modules for $\mathcal{G}^u$ and $\mathcal{G}^d$ that play the role of self-attention. Experiments show that our unrolled networks achieve competitive traffic forecast performance as state-of-the-art prediction schemes, while reducing parameter counts drastically.
format Preprint
id arxiv_https___arxiv_org_abs_2505_13102
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Lightweight and Interpretable Transformer via Mixed Graph Algorithm Unrolling for Traffic Forecast
Qi, Ji
Do, Tam Thuc
Liu, Mingxiao
Pan, Zhuoshi
Li, Yuzhe
Cheung, Gene
Zhao, H. Vicky
Machine Learning
Artificial Intelligence
Signal Processing
Unlike conventional "black-box" transformers with classical self-attention mechanism, we build a lightweight and interpretable transformer-like neural net by unrolling a mixed-graph-based optimization algorithm to forecast traffic with spatial and temporal dimensions. We construct two graphs: an undirected graph $\mathcal{G}^u$ capturing spatial correlations across geography, and a directed graph $\mathcal{G}^d$ capturing sequential relationships over time. We predict future samples of signal $\mathbf{x}$, assuming it is "smooth" with respect to both $\mathcal{G}^u$ and $\mathcal{G}^d$, where we design new $\ell_2$ and $\ell_1$-norm variational terms to quantify and promote signal smoothness (low-frequency reconstruction) on a directed graph. We design an iterative algorithm based on alternating direction method of multipliers (ADMM), and unroll it into a feed-forward network for data-driven parameter learning. We periodically insert graph learning modules for $\mathcal{G}^u$ and $\mathcal{G}^d$ that play the role of self-attention. Experiments show that our unrolled networks achieve competitive traffic forecast performance as state-of-the-art prediction schemes, while reducing parameter counts drastically.
title Lightweight and Interpretable Transformer via Mixed Graph Algorithm Unrolling for Traffic Forecast
topic Machine Learning
Artificial Intelligence
Signal Processing
url https://arxiv.org/abs/2505.13102