Spatial-Temporal Generative Adversarial Learning

Zhang, Yingxue

Etd

Spatial-Temporal Generative Adversarial Learning

Public

With the development of sensing and communication technologies, spatial-temporal big data has been widely generated and used in urban life, which helps to solve many problems related to smart cities, public safety, sustainability and business. However, it is challenging to deal with the spatial-temporal big data analytics problems (e.g., urban traffic estimation), because the data contains complex spatial-temporal dependencies, and is highly related to many other complicated factors. Given the large amount of spatial-temporal urban data, an important problem is how to successfully extract the complex spatial-temporal dependencies to solve diverse urban problems. In this dissertation, we will present an overview of my work that solves the spatial-temporal big data analytics problems in a deep generative adversarial perspective, and we will mainly introduce the spatial-temporal generative adversarial learning and its urban applications from the following 4 different perspectives. 1. The Conditional Urban Traffic Estimation problem aims to estimate the resulting traffic status prior to the urban construction plan. This problem is of great importance to urban development and transportation management, yet is very challenging due to the complex spatial-temporal dependencies and the relations between traffic and diverse urban conditions. To tackle this problem, we propose three different generative adversarial networks including TrafficGAN, Curb-GAN and C3-GAN. 2. Conventional methods for conditional traffic estimation usually focus on supervised settings, which require a large amount of labeled training data. However, in many urban planning applications, the large amount of traffic data in a new city can be hard or impossible to acquire. To tackle the conditional traffic estimation problem in data scarcity situations, we propose a novel STrans-GAN to extracts knowledge from multiple source cities to improve the estimation accuracy and transfer stability. 3. This topic mainly focuses on the problem of human urban strategy analysis. This problem is hard to solve due to two major challenges: (1) data scarcity and (2) data heterogeneity. To solve the human urban strategy analysis problem in case of data scarcity and data heterogeneity, we design a novel learning paradigm --- Spatial-Temporal Meta-GAIL (STM-GAIL), which can successfully learn diverse human urban strategies from heterogeneous human-generated spatial-temporal urban data. 4. Predicting traffic dynamics is of great importance to urban development and transportation management. However, it is very challenging to solve this problem due to spatial-temporal dependencies and traffic uncertainties. In this topic, we solve the traffic dynamics prediction problem from Bayesian meta-learning perspective and propose a novel continuous spatial-temporal meta-learner (cST-ML).

Creator