[论文解读] D-GAN: Deep Generative Adversarial Nets for Spatio-Temporal Prediction

Divya Saxena, Jiannong Cao|arXiv (Cornell University)|Jul 19, 2019

Traffic Prediction and Management Techniques参考文献 21被引用 27

一句话总结

本文提出D-GAN，一种用于时空预测的新型深度生成对抗网络，通过无监督方式联合学习潜在表征并执行变分推断。通过建模复杂、非线性的时空依赖关系，并利用融合模块整合外部因素，D-GAN在真实世界出租车需求和交通流量数据集上实现了最先进性能，优于传统方法和深度学习基线模型。

ABSTRACT

Spatio-temporal (ST) data for urban applications, such as taxi demand, traffic flow, regional rainfall is inherently stochastic and unpredictable. Recently, deep learning based ST prediction models are proposed to learn the ST characteristics of data. However, it is still very challenging (1) to adequately learn the complex and non-linear ST relationships; (2) to model the high variations in the ST data volumes as it is inherently dynamic, changing over time (i.e., irregular) and highly influenced by many external factors, such as adverse weather, accidents, traffic control, PoI, etc.; and (3) as there can be many complicated external factors that can affect the accuracy and it is impossible to list them explicitly. To handle the aforementioned issues, in this paper, we propose a novel deep generative adversarial network based model (named, D-GAN) for more accurate ST prediction by implicitly learning ST feature representations in an unsupervised manner. D-GAN adopts a GAN-based structure and jointly learns generation and variational inference of data. More specifically, D-GAN consists of two major parts: (1) a deep ST feature learning network to model the ST correlations and semantic variations, and underlying factors of variations and irregularity in the data through the implicit distribution modelling; (2) a fusion module to incorporate external factors for reaching a better inference. To the best our knowledge, no prior work studies ST prediction problem via deep implicit generative model and in an unsupervised manner. Extensive experiments performed on two real-world datasets show that D-GAN achieves more accurate results than traditional as well as deep learning based ST prediction methods.

研究动机与目标

• 解决在存在复杂非线性依赖关系和大量未列出的外部因素情况下，对高度不规则、动态的时空数据进行预测的挑战。
• 开发一种深度生成模型，隐式学习时空特征表征，而无需显式标注或枚举外部影响因素。
• 在无监督框架中联合优化生成与推断，以提升真实城市应用中的预测准确性。
• 通过隐式分布学习建模时空数据中的语义变化和不规则性。
• 通过专用融合模块将天气、交通管制和兴趣点（PoI）等外部因素整合到预测框架中。

提出的方法

• D-GAN采用基于GAN的架构，通过联合训练生成器和判别器来学习真实时空数据分布。
• 深层时空特征学习网络通过隐式建模数据分布，捕捉数据中的复杂相关性和潜在变化。
• 变分推断组件使模型能够在无真实标签的情况下，从观测数据中推断潜在表征。
• 融合模块将外部上下文因素（如天气、PoI、交通管制）整合到潜在空间中，以提升预测性能。
• 模型以端到端方式无监督训练，利用对抗损失和重建损失联合优化生成与推断。
• 生成器生成未来的时空序列，而判别器则区分真实序列与生成序列，从而鼓励生成更真实且多样的预测结果。

实验结果

研究问题

• RQ1 深度生成对抗网络是否能有效建模城市数据中复杂、非线性的时空依赖关系，而无需显式标注？
• RQ2 无监督模型在多大程度上能学习到捕捉动态时空序列中语义变化与不规则性的潜在表征？
• RQ3 整合外部上下文因素在多大程度上能提升时空预测的准确性？
• RQ4 生成与变分推断的联合优化是否能带来优于传统深度学习模型的预测性能？
• RQ5 在真实世界时空数据集上，D-GAN与最先进方法相比，在准确性和鲁棒性方面表现如何？

主要发现

• D-GAN在出租车需求和交通流量数据集上均优于传统统计模型和深度学习基线模型，实现了更优的预测精度。
• 该模型在多个时间预测范围内显著降低了预测误差（以RMSE衡量），表现出对数据不规则性的强鲁棒性。
• 通过融合模块整合外部因素后，预测性能得到可测量的提升，尤其在高变异性条件下更为明显。
• 无监督训练范式使模型无需标注数据或显式建模所有外部因素即可实现良好泛化能力。
• 消融实验确认，生成与推断两部分均对最终性能有显著贡献，验证了联合学习方法的有效性。
• 定性分析与判别器评估均证实，该模型能够生成多样化且逼真的未来序列。