|Title:||Urban computing with building data : an applied transfer learning approach|
|Advisors:||Wang, Dan (COMP)|
|Subject:||Hong Kong Polytechnic University -- Dissertations|
City planning -- Data processing
Human settlements -- Data processing
Municipal services -- Data processing
|Department:||Department of Computing|
|Pages:||xxii, 130 pages : color illustrations|
|Abstract:||Data mining and machine learning technologies have already achieved significant interest for urban computing. Early successes include the policing in New York City and monitoring the traffic status in Singapore. Above all, when urban data is used properly, urban computing can help tackle the challenges cities faced like pollution, energy, transportation problems. Thus, it is beneficial for people, environment and cities. Much recent progress in urban computing has been fuelled by the explosive growth in the amount and diversity of data available, and the computational resources needed to crunch through the data. The issue of data scarcity becomes more harmful especially when the distribution changes in urban computing systems, most statistical models need to be rebuilt from scratch using newly collected training data. However, in many real-world urban computing applications, it is expensive or impossible to collect the needed training data to build or rebuild the models. This begs the question of whether machine learning systems necessarily need large amounts of data to solve a task well. In urban computing, it would be beneficial to reduce the need and effort to collect and even recollect the training data. In this era of big data, as many different kinds of data have been collected, it is natural to ask whether we can take advantage of some other data to facilitate effective urban computing. This is the motivation behind this thesis on using big data collected from buildings for urban applications. The intuition is that the daily working and living patterns of people in the buildings may contribute to changes in the urban areas, e.g., in terms of traffic, population and energy footprint. The problem is then how to learn to learn efficiently, by making use of sim-ilarities among different domains, e.g., the built environment and urban scenarios. For this question, exciting recent practises, under the banners of meta-learning, life-long learning, learning to learn, multitask learning etc, lead us to the common answer of leveraging transfer learning. Transfer learning is to exposure to existing source domains and allows the system to learn a rich prior knowledge about the world in which tasks are sampled from, and it is with rich world knowledge that the system is able to solve new tasks efficiently. This is an active, vibrant and diverse area of research, with many different approaches proposed recently. This thesis describes three settings of transfer learning and the corresponding case studies in this direction that I have been involved in during my PhD study. The first study is on traffic prediction, which is an important application of tremendous practical value. In this case, we transfer samples from the domain of building CO2 to the domain of traffic speed, where the source domain and the target domain are shown to be related. As a case study of traffic prediction, in order to reduce the requirement of traffic data, we report a novel and interesting approach of warped inductive instance transfer using building data in Kowloon, Hong Kong, one of the densest urban areas in the world. The second study is on Urban Mobility Models (UMMs), which are fundamental tools for estimating the population in urban sites and their spatial movements over time. They have great value for such applications as managing the resources of cellular networks, predicting traffic congestion, and city planning. In this study, in order to reduce the requirement of mobility data, we transfer data instance from the domain of building occupancy to the domain of traffic mobility, where we show that the two domains have some common characteristics. As a case study of semi-absorbing UMM, we proposed a novel crossed inductive instance transfer developed with building data. The third study is on chiller performance prediction, which is significant for the core problem of building operation to reduce HVAC electricity consumption. In this study, we confronted with the data scarcity issue in quite a few target domains, where both the source and the target features lie in the same space but with different distributions. As a case study of chiller performance prediction, we present a novel grouped inductive instance transfer to solve the data scarcity issue. In summary, we propose three new methods of inductive instance transfer to generate hidden insights and enable intelligent decision-making without too much effort on data collection. The proposed methods are leveraged in important urban computing applications covering three possible cases in inductive instance transfer in terms of domain. We identify the requirements and address the challenges therein, providing effective frameworks and solutions for practitioners as well as offering useful insights for future research.|
|Rights:||All rights reserved|
Files in This Item:
|991022249145403411.pdf||For All Users||5.99 MB||Adobe PDF||View/Open|
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item: