Projects

Research & Publications

Authors:
Tsang, Y. P., Choy, K .L., Wu, C.H., Ho, G.T.S.     

Abstract:
Effective deployment of the emerging environmental sensor network in environmental mapping has become essential in numerous industrial applications. The essential factors for deployment include cost, coverage, connectivity, airflow of heating, ventilation, and air conditioning, system lifetime, and fault tolerance. In this letter, a three-stage deployment scheme is proposed to formulate the above-mentioned considerations, and the fuzzy temperature window is established to adjust sensor activation times over various ambient temperatures. To optimize the deployment effectively, a multi-response Taguchi-guided k-means clustering is proposed to embed in the genetic algorithm, where an improved set of the initial population is formulated and system parameters are optimized. Therefore, the computational time for repeated deployment is shortened, while the solution convergence can be improved.

Authors:
Ho, G.T.S. , Tsang, Y.P., Wu, C.H., Wong, W.H., Choy, K.L.            

Abstract:
In digital and green city initiatives, smart mobility is a key aspect of developing smart cities and it is important for built-up areas worldwide. Double-parking and busy roadside activities such as frequent loading and unloading of trucks, have a negative impact on traffic situations, especially in cities with high transportation density. Hence, a real-time internet of things (IoT)-based system for surveillance of roadside loading and unloading bays is needed. In this paper, a fully integrated solution is developed by equipping high-definition smart cameras with wireless communication for traffic surveillance. Henceforth, this system is referred to as a computer vision-based roadside occupation surveillance system (CVROSS). Through a vision-based network, real-time roadside traffic images, such as images of loading or unloading activities, are captured automatically. By making use of the collected data, decision support on roadside occupancy and vacancy can be evaluated by means of fuzzy logic and visualized for users, thus enhancing the transparency of roadside activities. The CVROSS was designed and tested in Hong Kong to validate the accuracy of parking-gap estimation and system performance, aiming at facilitating traffic and fleet management for smart mobility.

Authors:
Mo, D. Y., Ng, S. C. H., Tai, David.

Abstract:
This study demonstrates how NetApp, a data storage system provider, used Six Sigma to solve the service parts inventory problem in its multiechelon logistics network, which its inventory management system was unable to fix. The nonstationary demand for service parts created a blind spot for the system, thus hampering NetApp’s contractual commitment to customers of an almost 100% fill rate (FR) for replacing service parts. Constant customer complaints because of FRs that were less than 100% caused NetApp to improve the performance of its service parts replenishment and order fulfillment processes. By following the Six Sigma approach and using the associated qualitative and quantitative tools, the company worked systemically to identify the major causes of insufficient stock and systematically corrected the problem. NetApp formulated a cost-effective inventory solution for its inventory planning system, which resulted in a 10% decrease in the ratio of inventory to revenue and an FR increase from 99.1% to 99.6%. The standard deviation of the replenishment lead time also declined from 4.97 to 1.87 days, implying that the variation of the replenishment lead time was greatly reduced. The Six Sigma process, therefore, provided new insights and a new approach to enable NetApp to manage its inventory planning process.

Authors:
Lo, W. H., Lam, B. S. Y., Cheung , M. F.

Abstract:
This article examines the news framing of the 2017 Hong Kong Chief Executive election using a big data analysis approach. Analyses of intermedia framing of over 370,000 articles and comments are conducted including news published in over 30 Chinese press media, four prominent Chinese online press media, and posts published on three candidates’ Facebook pages within the election period. The study contributes to the literature by examining the rarely discussed role of intermedia news framing, especially the relationship between legacy print media, online alternative news media, and audience comments on candidates’ social network sites. The data analysis provides evidence that audiences’ comments on candidates’ Facebook pages influenced legacy news coverage and online alternative news coverage. However, this study suggests that legacy news media and comments on Facebook do not necessarily have a reciprocal relationship. The implication of the findings and limitations are discussed.

Authors:
Lam, B. S. Y., Choy , S. K.

Abstract:
Different versions of principal component analysis (PCA) have been widely used to extract important information for image recognition and image clustering problems. However, owing to the presence of outliers, this remains challenging. This paper proposes a new PCA methodology based on a novel discovery that the widely used -PCA is equivalent to a two-groups -means clustering model. The projection vector of the -PCA is the vector difference between the two cluster centers estimated by the clustering model. In theory, this vector difference provides inter-cluster information, which is beneficial for distinguishing data objects from different classes. However, the performance of -PCA is not comparable with the state-of-the-art methods. This is because the -PCA can be sensitive to outliers, as the equivalent clustering model is not robust to outliers. To overcome this limitation, we introduce a trimming function to the clustering model and propose a trimmed-clustering based -PCA (TC-PCA). With this trimming set formulation, the TC-PCA is not sensitive to outliers. Besides, we mathematically prove the convergence of the proposed algorithm. Experimental results on image classification and clustering indicate that our proposed method outperforms the current state-of-the-art methods.

Authors:
Xu, L., Tang, M. L., Chen, Z.

Abstract:
In longitudinal data analysis, it is crucial to understand the dynamic of the covariance matrix of repeated measurements and correctly model it in order to achieve efficient estimators of the mean regression parameters. It is well known that any incorrect covariance matrices can result in inefficient estimators of the mean regression parameters. In this article, we propose an empirical likelihood based method which combines the advantages of different dynamic covariance modeling approaches. The effectiveness of the proposed approach is demonstrated by an anesthesiology dataset and some simulation studies.

Ongoing Projects

Principal Investigator:
Dr. Choy Siu-Kai

Abstract:
Image segmentation is a critical problem in computer vision for a wide variety of applications. Among the existing approaches, partial differential equations and variational methods have been extensively studied in the literature. Although most variational approaches use boundary and region information to segment natural and textural images with remarkable success, we note that most of the existing methods only consider simple information/features extracted from a particular image domain (e.g., grey level features in the spatial domain) to characterise image regions. However, such information/features are not informative enough to segment complex images. In the proposed project, we will investigate a robust and effective variational segmentation algorithm to remedy the aforementioned difficulties for a wide range of applications. In particular, we will study a mathematical optimisation framework that integrates the bit-plane-dependence probability models, which are used to characterise local region information extracted from various image domains, with the fuzzy region competition for image segmentation. We will also study the mathematical theory for the segmentation algorithm. The proposed segmentation method will be assessed by extensive and comparative experiments using complex natural and textural images.

Principal Investigator:
Dr. Choy Siu-Kai

Abstract:
Image segmentation is a challenging problem in computer vision and has a wide variety of applications in various fields such as pattern recognition and medical imaging. One of the main approaches to this problem is to perform superpixel segmentation followed by a graph-based methodology to achieve image segmentation. Crucial to the successful image segmentation using this method is the superpixel generation algorithm and superpixel partitioning algorithm. Existing superpixel generation algorithms have various priorities and place emphasis on boundary adherence, superpixel regularity, computational complexity, etc, but normally do not perform well in all of the above simultaneously. Superpixel partitioning algorithms are typically based on graph-based approaches and could have high computational costs, which makes them inefficient in practical contexts. In the proposed project, we will investigate a fast and effective unsupervised fuzzy superpixel-based image segmentation algorithm to remedy the aforementioned difficulties for a wide range of applications. In particular, we will study the combined use of a novel fuzzy clustering-based superpixel generation technique and fuzzy graph-theoretic superpixel partitioning approach for image segmentation applications. The proposed segmentation method will be assessed by extensive comparative experiments using complex natural and textural images.

Principal Investigator:
Dr. Lam Shu-Yan

Abstract:
Supervised learning problems infer functions from labelled training data. Learning lower dimensional subspaces in supervised learning problems is important in applications such as human action recognition, face recognition and object recognition. Dimensionality reduction is performed to remove noise from the data and simplify data analysis. Linear Discriminant Analysis (LDA) and its variants have been shown to be suitable for handling data structures in linear, quadratic and highly non-linear forms. However, conventional LDA formulations suffer from two major limitations. Firstly, they use arithmetic means to represent the class centroids of the input data. However, the arithmetic mean has been shown to not effectively represent these data, especially with data that contains heavy noise and outliers. Secondly, it is difficult to show statistically that the learnt projection vectors are effective in the presence of heavy noise and outliers. Hence, conventional LDA fails to determine the most representative features from the input data.

In the proposed project, we aim to develop a new class of dimensionality reduction techniques for labelled data that can overcome the major limitations of conventional LDA techniques. The core idea is to formulate the dimensionality reduction problem as a set of clustering problems. The novelty of the proposed approach is that unsupervised clustering problems can effectively learn the subspace of the supervised learning problem. Locating effective centroids has been well-studied in clustering research. Furthermore, well-developed theories can be used to analyse the sensitivities of these methods in the presence of heavy noise and outliers. If successful, the proposed study will significantly increase the performance of dimensionality reduction for labelled data using clustering, which will fundamentally improve the way in which useful information can be extracted in many real-world applications.

Principal Investigator:
Prof. Tang Man-Lai

Abstract:
One of the most important challenges in modern survey measurement is the elicitation of truthful answers to sensitive questions about behavior and attitudes (e.g., abortion, illegal drug use and racial prejudice). It has long been well known that accessing information regarding a sensitive characteristic in a population usually induces two notorious issues, namely non-response bias (i.e., respondents refuse to collaborate in the fear of the protection of their confidentiality) and response bias (e.g., respondents answer the sensitive questions but give false answers), which usually induce estimate’s efficiency loss, inflated sampling variance, and biased estimates. Therefore, techniques that guarantee anonymity, minimize the respondents’ feelings of jeopardy, and encourage honest answers are of great demand. In this project, we propose several practical generalizations for the famous item count techniques for sensitive survey questions.

Poisson ICT has recently been developed to overcome the shortages of the conventional item count techniques (ICTs) by replacing the list of independent innocuous (binary) statements by a single innocuous (Poisson) statement. Despite various attractive advantages, Poisson ICT still possesses some limitations. First, it is assumed that respondents will comply with the survey design. Second, it is assumed that the outcome of the innocuous statement follows the less practical Poisson distribution. Third, no regression model has been developed for binary sensitive outcomes.

In this proposal, we plan to

(i) (New Poisson ICT with Non-Compliance) Develop a new Poisson ICT that takes the non-compliance from the respondents into consideration;

(ii) (New Inflated-Zero Poisson ICT) Develop a new Poisson-type ICT that allows the outcome of the innocuous statement follows the more realistic inflated-zero Poisson distribution; and

(iii) (Regression Modeling with Sensitive Outcome) Develop a regression model for binary sensitive outcomes.

Principal Investigator:
Prof. Tang Man-Lai

Abstract:
High dimensional data analysis has become increasingly frequent and important in diverse fields; for example, genomics, health sciences, economics and machine learning. Model selection plays a pivotal role in contemporary scientific discoveries. There have been a large body of works on model selection for complete data. However, complete data are often not available for every subject due to many reasons, including the unavailability of covariate measurements and loss of data. The literature on model selection for high dimensional data in the presence of missing or incomplete values is relatively sparse. Therefore, efficient methods and algorithms for model selection with incomplete data are of great research interest and practical demand.

For model selection, the information criteria (e.g., the Akaike information criterion and the Bayesian information criterion) is commonly applied, and it can be easily incorporated with the famous EM algorithm in the presence of missing values. Generalized EM algorithm has also been developed to update the model and the parameter under the model in each iteration. It performs Expectation step and Model Selection Step alternately, converges globally, and yields a consistent model in model selection. However, it may not always be numerically feasible to perform Model Selection Step, especially for high dimensional data. Therefore, a new method for model selection with high dimensional incomplete data is greatly desirable. Our proposed algorithm in this project will hopefully yield a consistent model in general missing data patterns and have numerical convergences. Moreover, our proposed method is expected to perform efficiently variable selection in linear regression, generalized linear models and model selection of graphical models.

Due to the convenience of its implementation by using standard software modules, multiple imputation is arguably the most widely used approach for handling missing data. It is straightforward to apply an existing model selection method to each imputed dataset. However, it is challenging to combine results on model selection across imputed data sets in a principled framework. To overcome the challenge, many advanced techniques are developed for variable selection problem, such as the group lasso penalty to merged data sets of all imputations, the strategy of stability selection within bootstrap imputation, and random lasso combined with multiple imputation. These techniques are feasible for high-dimensional data with complex missing patterns and have achieved good performance in simulation studies and real data analyses. However, as far as we know, it is very surprising that there is no imputation method for graphical models. An imputation-based method for graphical model selection is greatly desirable. In this project, we investigate bootstrap multiple imputation with stability selection. We expect the proposed method can deal with general missing data patterns.

Principal Investigator:
Dr. Liu Hai

Abstract:
Robotics technologies are advancing rapidly. Groups of robots have been developed that can communicate with each other using wireless transmissions and form robot swarms. Applications of these swarms include surveillance, search and rescue, mining, agricultural foraging, autonomous military units and distributed sensing in micromachinery or human bodies. For example, swarm robots can be sent into the places that are too dangerous for human workers and detect life-signs via infrared sensors. In all of these applications, a group of self-propelled robots move in a cohesive way (i.e., connectivity is preserved during these movements). Such behavior is usually referred to as collective motion. This research aims to design self-adaptive collective motion algorithms for swarm robots in 3D space. The algorithms are expected to be self-adaptive in the sense that robots will be able to dynamically determine proper moving parameters, based on their environments and statuses. Using the proposed collective motion algorithms, robots will be able to move along a pre-planned path from a source to a destination while satisfying the following requirements. 1) The robots will use only one-hop neighbor information. 2) The robots will maintain connectivity of the network topology for information exchange. 3) The robots will maintain a desired neighboring distance. 4) The robots will be capable of bypassing obstacles without partitioning the robot swarm (i.e., member loss). We will develop collective motion algorithms for the following three cases: 1) no obstacles and no leader, 2) no obstacles with a leader; and 3) with obstacles (with and without a leader). We will conduct extensive experiments in testbed robots to examine the performance of the algorithms in practical applications.

Principal Investigator:
Dr. Ho To-Sum

Abstract:
The blooming of e-commerce in the past decade has not only brought significant economic growth to the e-retailers, but also new opportunities and challenges to the logistics industry. To seize the opportunities arising from the emerging e-commerce logistics in Hong Kong, logistics service providers (LSPs) are forced to take on new roles and adjust their operations to fulfill the dynamic customer demand. This research aims to develop a Blockchain-based E-Commerce Analytics Model, integrating blockchain technology and the machine learning algorithm for managing data across the supply chains and predicting dynamic e-commerce order demand.

This research enables industry practitioners, especially LSPs and e-retailers, to plan ahead for the subsequent e-commerce operations. From the perspective of LSPs, the prediction model allows the firm to realize the e-commerce order arrival patterns, enabling flexible re-allocation of the right amount of resources in real time to deal with the hour-to-hour fluctuating arrival of orders in distribution centers. From the perspective of a retailer, the generic prediction model allows the firm to predict, for example, the sales volume among various e-commerce sales channels, the sales volume from different customer segments, and the e-commerce sales performance of different product categories. By tackling the unpredictability of demand in the e-commerce business environment, this research contributes to an effective decision support strategy for logistics `operations planning, hence, enhancing e-commerce logistics competence in Hong Kong.

Principal Investigator:
Dr. Mo Yiu-Wing

Abstract:
Given the current ageing population and limited social welfare expenditure, scholars are renewing their interests in how community organisations can operate to sustainably serve various needs of people with travel inconvenience in society. This research aims to design flexible vehicle management systems that enhance the management of various paratransit services through better system design and optimisation of vehicle resources.

The study scope of paratransit services includes schedule route, dial-a-ride, feeder and pooled dial-a-ride. Users who require those services have different expectations for travelling times, prices, service frequencies as well as pick-up and drop-off locations. This variety of service requirements poses numerous new challenges for community organisations to sustain paratransit services. Hence, it is essential to innovate options for a holistic approach to coordinating various types of service in a common sharing platform, which meet people’s diverse needs in a more efficient way. We expect the outcomes of this research would support the policy review and operational improvements for community organisations.

Principal Investigator:
Dr. Mo Yiu-Wing

Abstract:
With the advanced logistics developments in recent decades, various manufacturers are able to profit from the spare parts service for systems maintenance and to enhance product sustainability by managing the express delivery and the reverse logistics. These advanced logistics developments have driven the evolution of traditional spare parts management into a new service model. Apart from the on-site spare parts management, manufacturers and authorised service providers must offer more customised services and the collection of repairable items from users in the reverse logistics process. However, these evolutionary service requirements introduce procedural complexities and extends the service scope.

In this research, we aim to optimise the process of service parts management through a holistic and adaptive approach. The whole process scope includes logistics network design, inventory and warehouse management, and reverse logistics operations. To identify the numerous factors and parameters during the process optimisation, we will start by standardising a generic process flow of service parts operations that align with companies’ strategic objectives. Then, we will perform data collection to investigate the effects of these factors and their correlations. After identifying the critical factors, we will formulate them into a generic decision model for deriving optimal adaptive policies with a data-driven process control mechanism. A simulation platform will be developed to verify and monitor the proposed solutions. The performance of the optimal adaptive policies will be finally benchmarked with the optimal static policy, which is commonly applied in various industries. These results will provide effective guidelines for the implementation of adaptive process optimisation of service parts operations.

Principal Investigator:
Dr. Wong Siu-kuen Ricky

Abstract:
Past studies on negotiation strategy have emphasised the benefits of different compliance techniques, for example, door-in-the-face, foot-in-the-door, the low-balling techniques, anchoring effect, etc. A growing body of research has shown how negotiators using compliance tactic may obtain better negotiated outcomes. Undoubtedly, the use of these tactics is beneficial when there involves only a one-off negotiation. Now we have seen that many opportunities for negotiation training are available at universities and corporate training courses. And, in a real-life setting, it is often that negotiators involve in repeated negotiation. Coupling this with people’s knowledge in negotiation tactics, it is contentious that the use of compliance tactic is beneficial in the longer run. The adverse effects of compliance tactic have been neglected in research on negotiation. A more thorough understanding of the potential costs resulting from the use of compliance tactics is important for negotiators or practitioners to make an informed decision.

Upcoming Projects

Principal Investigator: Prof. Tang Man-Lai