Qing Li

Lands Department of The Government of the Hong Kong Special Administrative Region, Hong Kong, Hong Kong

Are you Qing Li?

Claim your profile

Publications (272)62.22 Total impact

  • [show abstract] [hide abstract]
    ABSTRACT: The skyline operator has been extensively explored in the literature, and most of the existing approaches assume that all dimensions are available for all data items. However, many practical applications such as sensor networks, decision making, and location-based services, may involve incomplete data items, i.e., some dimensional values are missing, due to the device failure or the privacy preservation. This paper is the first, to our knowledge, study of k-skyband (kSB) query processing on incomplete data, where multi-dimensional data items are missing some values of their dimensions. We formalize the problem, and then present two efficient algorithms for processing it. Our methods introduce some novel concepts including expired skyline, shadow skyline, and thickness warehouse, in order to boost the search performance. As a second step, we extend our techniques to tackle constrained skyline (CS) and group-by skyline (GBS) queries over incomplete data. Extensive experiments with both real and synthetic data sets demonstrate the effectiveness and efficiency of our proposed algorithms under various experimental settings.
    Expert Systems with Applications 01/2014; 41(10):4959–4974. · 1.85 Impact Factor
  • [show abstract] [hide abstract]
    ABSTRACT: Rare category discovery aims at identifying unlabeled data examples of rare categories in a given data set. The existing approaches to rare category discovery often need a certain number of labeled data examples as the training set, which are usually difficult and expensive to acquire in practice. To save the cost however, if these methods only use a small training set, their accuracy may not be satisfactory for real applications. In this paper, for the first time, we propose the concept of rare category exploration, aiming to discover all data examples of a rare category from a seed (which is a labeled data example of this rare category) instead of from a training set. To this end, we present an approach known as the FRANK algorithm which transforms rare category exploration to local community detection from a seed in a kNN (k-nearest neighbors) graph with an automatically selected k value. Extensive experimental results on real data sets verify the effectiveness and efficiency of our FRANK algorithm.
    Expert Systems with Applications 01/2014; 41(9):4197–4210. · 1.85 Impact Factor
  • [show abstract] [hide abstract]
    ABSTRACT: This paper presents an efficient and robust content-based large medical image retrieval method in mobile Cloud computing environment, called the Mirc. The whole query process of the Mirc is composed of three steps. First, when a clinical user submits a query image Iq, a parallel image set reduction process is conducted at a master node. Then the candidate images are transferred to the slave nodes for a refinement process to obtain the answer set. The answer set is finally transferred to the query node. The proposed method including an priority-based robust image block transmission scheme is specifically designed for solving the instability and the heterogeneity of the mobile cloud environment, and an index-support image set reduction algorithm is introduced for reducing the data transfer cost involved. We also propose a content-aware and bandwidth-conscious multi-resolution-based image data replica selection method and a correlated data caching algorithm to further improve the query performance. The experimental results show that the performance of our approach is both efficient and effective, minimizing the response time by decreasing the network transfer cost while increasing the parallelism of I/O and CPU.
    Information Sciences. 01/2014; 263:60–86.
  • [show abstract] [hide abstract]
    ABSTRACT: Langerhans cell histiocytosis (LCH) and Erdheim-Chester disease (ECD) share similar clinical features and mechanisms. In very rare circumstances, the two diseases coexist in the same patient. Here we report such a patient, who was first diagnosed with Hand-Schüller-Christian disease (HSC), a type of LCH. Several years later, the patient presented with severe exophthalmos and osteosclerosis on radiograph. New biopsy revealed ECD. We also analyze 54 cases of LCH and 6 cases of ECD diagnosed in our hospital, as well as their progression during a follow-up period of 8 years. In five cases of HSC (9.3% of LCH), a triad of central diabetes insipidus, hyperprolactinemia, and pituitary stalk thickening on magnetic resonance imaging (MRI) preceded the typical bone lesions by 4-9 years. In addition, LCH was featured as elevated plasma alkaline phosphatase (ALP), which was normal in ECD. Combined with a literature review, several features are summarized to differentiate ECD from HSC. In patients with diabetes insipidus, concomitant hyperprolactinemia and pituitary stalk thickening on MRI indicate a possible HSC. Additionally, if osteosclerosis is observed in a patient with LCH, the coexistence of ECD should be considered.
    The Oncologist 01/2013; · 4.10 Impact Factor
  • [show abstract] [hide abstract]
    ABSTRACT: Text representation is one of the most fundamental works in text comprehension, processing, and search. Various works have been proposed to mine the semantics in texts and then to represent them. However, most of them only focus on how to mine semantics from the text itself while the background knowledge, which is very important to text understanding, is not taken into consideration. In this paper, on the basis of human cognitive process, we propose a multi-level text representation model within background knowledge, called TRMBK. It is composed of three levels, which are machine surface code (MSC), machine text base (MTB) and machine situational model (MSM). All of the three are able to be automatically constructed to acquire semantics both inside and outside of the text. Simultaneously, we also propose a method to automatically establish background knowledge and offer supports for the current text comprehension. Finally, experiments and comparisons have been presented to show the better performance of TRMBK.
    Cognitive Informatics & Cognitive Computing (ICCI*CC), 2013 12th IEEE International Conference on; 01/2013
  • Source
    [show abstract] [hide abstract]
    ABSTRACT: The k -nearest-neighbor ( k -NN) query is one of the most popular spatial query types for location-based services (LBS). In this paper, we focus on k -NN queries in time-dependent road networks, where the travel time between two locations may vary significantly at different time of the day. In practice, it is costly for a LBS provider to collect real-time traffic data from vehicles or roadside sensors to compute the best route from a user to a spatial object of interest in terms of the travel time. Thus, we design SMashQ, a server-side spatial mashup framework that enables a database server to efficiently evaluate k -NN queries using the route information and travel time accessed from an external Web mapping service, e.g., Microsoft Bing Maps. Due to the expensive cost and limitations of retrieving such external information, we propose three shared execution optimizations for SMashQ, namely, object grouping , direction sharing , and user grouping , to reduce the number of external Web mapping requests and provide highly accurate query answers. We evaluate SMashQ using Microsoft Bing Maps, a real road network, real data sets, and a synthetic data set. Experimental results show that SMashQ is efficient and capable of producing highly accurate query answers.
    Distributed and Parallel Databases 09/2012; · 0.81 Impact Factor
  • [show abstract] [hide abstract]
    ABSTRACT: Some nondiabetic trauma patients with hyperglycemia have been found to have occult diabetes mellitus (ODM). We studied whether glycated albumin (GA) was an effective tool for detecting ODM in orthopedic trauma patients with elevated glucose levels. A cross-sectional, sequential case series study of adult patients presenting to the Orthopedic Trauma Center between September 2009 and March 2010 with new limb fractures was performed. Hemoglobin A1c (HbA1c) and GA levels were measured in hyperglycemic patients with no prior diabetes mellitus. A receiver operating characteristic curve was plotted to examine the sensitivity, specificity, and positive and negative predictive values of GA in identifying occult diabetes in hospitalized patients with acute hyperglycemia. A total of 2,058 trauma patients were screened and 399 patients (19.4%) with no known diabetes mellitus were noted to be hyperglycemic. Of these 399 patients, 38.3% (n = 153) had ODM according to the HbA1c diagnosis cutoff point. GA level was strongly correlated with HbA1c using Pearson's correlation analysis (r = 0.887, p < 0.01). Using logistic regression analysis, GA (odds ratio [OR] = 1.60, p < 0.001) and fasting plasma glucose (OR = 1.974, p < 0.001) were identified as significant risk factors for the diagnosis of ODM. Receiver operating characteristic curve analysis showed that a GA value of 17.5% gave an optimal sensitivity of 73.20% and specificity of 99.12% for distinguishing ODM from stress-induced hyperglycemia. Almost 40% of nondiabetic orthopedic trauma patients presenting with hyperglycemia were found to have ODM. A GA value of 17.5%, the optimal cutoff point, could distinguish between ODM and stress-induced hyperglycemia in Chinese orthopedic trauma subjects. II, diagnostic study.
    The journal of trauma and acute care surgery. 05/2012; 72(5):1369-74.
  • Source
    [show abstract] [hide abstract]
    ABSTRACT: Our previous study showed there was a gender difference in plasma lactate concentrations in subjects with type 2 diabetes. This study investigated the effect of sex hormone levels on plasma lactic acid (LA) levels in type 2 diabetes with and without metformin therapy. Fasting whole blood specimens of 392 type 2 diabetes patients treated with metformin (n=199) or not (n=193) were collected. LA was measured with an enzyme-electrode assay. Levels of sex hormones, including testosterone (T) and estradiol (E(2)), were measured with a chemiluminescence microparticle immunoassay. Spearman's or Pearson's correlation and logistic regression analysis were performed for the factors associated with LA. The LA level in the metformin group was significantly higher than that in the non-metformin group (1.26±0.43 vs. 1.14±0.49 mmol/L, P<0.001), and LA levels of females were significantly higher than those of males (P<0.001). LA concentrations were positively correlated with E(2) level but negatively correlated with metformin and T levels (P<0.01). The logistic regression analysis showed that gender, creatinine, E(2), metformin, and T were independent factors influencing lactate levels. Analysis of subgroups demonstrated that the LA concentrations increased with the elevation of E(2) level (P<0.05) but decreased with the rising of T level (P<0.05). Sex hormones play an important role on regulating plasma lactate levels in diabetes patients treated with metformin. E(2) up-regulates but T tend to down-regulate lactate levels.
    Diabetes Technology &amp Therapeutics 04/2012; 14(6):469-74. · 2.21 Impact Factor
  • [show abstract] [hide abstract]
    ABSTRACT: To explore the relationship between glycated albumin (GA) level and pancreatic β cell function in newly diagnosed type 2 diabetics. The subjects sought the confirmation of diabetes diagnosis or underwent diabetes screening tests in high-risk patients from January 2008 to October 2010. All of them underwent 75 g oral glucose tolerance test (OGTT) and insulin releasing test. The levels of GA and hemoglobin A1c (HbA1c) were analyzed by liquid enzymatic method and high performance liquid chromatography respectively. Homeostasis model assessment (HOMA) was used to evaluate the basal insulin resistance (HOMA-IR) and pancreatic β cell function (HOMA-β). ΔI30/ΔG30 was used to evaluate early-phase insulin secretion after a glucose load. (1) Among 500 type 2 diabetics according to the diagnostic criteria of WHO (1999), 279 were males and 221 were females. Average age was 56.3 ± 12.3, GA (21.1 ± 5.4)% and HbA1c (7.0 ± 1.3)%. (2) A significantly positive relationship was shown between HbA1c and GA (r = 0.691, P < 0.01). GA was also positively correlated with fasting plasma glucose (FPG), 0.5 hPG, 1 hPG, 2 hPG and 3 hPG after a glucose load of OGTT test (r = 0.511 - 0.627, P < 0.01). (3) GA was negatively correlated with body mass index (BMI) (r = -0.112, P < 0.01), HOMA-β (r = -0.350, P < 0.01) and ΔI30/ΔG30 (r = -0.263, P < 0.01). (4) Multivariant stepwise regression analysis showed that HbA1c, FPG, 3 hPG and ΔI30/ΔG30 were independent factors of GA level. Glycated albumin level is closely correlated with the function of early-phase insulin secretion in newly diagnosed type 2 diabetics.
    Zhonghua yi xue za zhi 03/2012; 92(12):814-7.
  • Source
    [show abstract] [hide abstract]
    ABSTRACT: To investigate a potential relationship between Solute carrier family 30 (zinc transporter) member 8 (SLC30A8) rs13266634 variant and efficacy of rosiglitazone or repaglinide in treating newly diagnosed Chinese type 2 diabetes patients. A total of 209 diabetic patients without any antihyperglycemic history were recruited and treated with repaglinide or rosiglitazone randomly for 48 weeks (104 and 105 patients, respectively). Anthropometric measurements and clinical laboratory tests were carried out before and after the treatment. An non-synonymous variant rs13266634 was genotyped by matrix-assisted laser desorption ionization-time of flight mass spectroscopy. Ninety-one patients in repaglinide group and ninety-three patients in rosiglitazone group completed the study. Δ value of homeostasis model assessment of beta cell function (HOMA-B) and Δ value of fasting proinsulin levels were statistically significant between three genotype groups (P=0.0149 and 0.0246, respectively) after rosiglitazone treatment. However, no genotype association was observed in the repaglinide or rosiglitazone group with other parameters. The SLC30A8 variant was associated with the efficacy of insulin sensitizer monotherapy on insulin secretion in patients with newly diagnosed type 2 diabetes mellitus in Shanghai, China.
    Biomedical and Environmental Sciences 02/2012; 25(1):23-9. · 1.15 Impact Factor
  • Source
    [show abstract] [hide abstract]
    ABSTRACT: To evaluate the prevalence of atherosclerosis detected by both carotid and lower extremity ultrasonography in hospitalized Chinese type 2 diabetic patients and to examine whether plaque formation in the carotid arteries could be an indicator of generalized atherosclerosis in type 2 diabetes mellitus. Totally, 709 hospitalized Chinese type 2 diabetic patients (men 357, women 352) aged from 18 to 88 years were included. Both carotid and lower extremity atherosclerosis were assessed by Doppler ultrasound. Atherosclerosis was defined as the presence of either the carotid or lower extremity plaque in any of the above-mentioned arteries segments. The prevalence of atherosclerosis was calculated, and the risk factors associated with atherosclerosis were evaluated using binary logistic regression. The prevalence of atherosclerosis was 81.23% in male and 77.56% in female type 2 diabetic patients, respectively. There was no significant difference in the prevalence of atherosclerosis in patients between the sexes. The prevalence of atherosclerosis was significantly higher in the lower extremity arteries than in the carotid arteries (73.91% and 44.43%, respectively, P<.001). Atherosclerosis was significantly associated with smoking, age, duration of diabetes, systolic blood pressure, total number of white blood cells, and mean carotid and femoral intima-media thickness (IMT). The prevalence of atherosclerosis was very high in Chinese inpatients with type 2 diabetes. Carotid atherosclerosis could not be an indicator of generalized atherosclerosis in type 2 diabetes. The combination of carotid and lower extremity ultrasound examination can significantly improve the detection of atherosclerosis in type 2 diabetes.
    Journal of diabetes and its complications 01/2012; 26(1):23-8. · 2.11 Impact Factor
  • [show abstract] [hide abstract]
    ABSTRACT: Faceted search on web pages needs exact facets. However, it is difficult to extract facets exactly from web pages because the web pages are unstructured and lack of facet information. Therefore, facet extraction is a key to faceted search. This paper proposed a method of extracting facets automatically from unstructured web pages to improve the faceted search on web. The Multidimensional Semantic Index (MDSI) of web pages is constructed by mining all kinds of semantic relations among the words from web pages, which creates a semantic-rich index for web pages. In MDSI, the differently dimensional semantic indexes are bridged by mining the semantic mapping between them. Based on the MDSI of web pages, the facets are extracted by analyzing semantic mapping relations in MDSI. To validate the effect of the proposed method, two datasets are constructed and the experimental results show that the proposed method is feasible and comparatively precise.
    Semantics, Knowledge and Grids (SKG), 2012 Eighth International Conference on; 01/2012
  • [show abstract] [hide abstract]
    ABSTRACT: The protein tyrosine phosphatase non-receptor 22 (PTPN22) gene encodes for lymphoid protein tyrosine phosphatase. Recent studies demonstrated the association between the +1858T, -1123G>C variants of PTPN22 gene and type 1 diabetes mellitus in Caucasian and Japanese populations. This study examined the relationship between the polymorphism of PTPN22 gene and latent autoimmune 1 diabetes in adults (LADA) in Chinese Hans. We studied 229 adult Chinese patients with LADA (LADA group) and 210 healthy volunteers (control group). The -1123G>C and +1858C>T polymorphisms of PTPN22 gene were determined by PCR-restriction fragment length polymorphism method. Further, genotypic/allelic frequencies and clinical characteristics were compared between two groups. There was a significant difference of frequencies of the -1123G>C polymorphism between LADA and control groups (OR = 1.99, 95% CI = 1.24-3.2; P = 0.001). However, no significant differences in the +1858C>T genotypic (CC, CT) and allelic (C, T) frequencies were found. Furthermore, the frequencies of the -1123 GC, CC genotype in male patients with LADA were significantly higher compared with male healthy volunteers (OR = 1.65, 95% CI = 1.21-2.26; P = 0.005). The analysis of covariance demonstrated no difference between glycosylated hemoglobin, body mass index, duration of diabetes, C-peptide, and GAD-Ab titer between the group carrying GC/CC and the group without allele C. In conclusion, the -1123G>C promoter polymorphism of PTPN22 gene, but not the +1858C>T variant, is associated with LADA in adult Chinese Hans.
    Cell biochemistry and biophysics 09/2011; 62(2):273-9. · 3.34 Impact Factor
  • [show abstract] [hide abstract]
    ABSTRACT: To assess the validity of combined detection of hemoglobin A1c (HbA1c) and glycated albumin (GA) in diabetic screening. A total of 1480 subjects at our out-patient department from March 2007 to December 2009. Those suspected of diabetes or at a high risk of diabetes were enrolled. The study population included 677 males and 803 females with a mean age of 52.7 years. All subjects received an oral glucose tolerance test (OGTT) after a 10-hour fasting. Glycated albumin (GA) and hemoglobin A1c (HbA1c) were measured with liquid enzyme method and high pressure liquid chromatography respectively. (1) According to World Health Organization diabetes diagnosis criteria, there were 562 subjects with normal glucose tolerance (NGT), 411 subjects with impaired glucose regulation (IGR) and 507 subjects with newly diagnosed diabetes mellitus (DM). The level of HbA1c and GA had a rising tendency among NGT, IGR and DM groups (P < 0.01). (2) Pearson correlation analysis demonstrated that HbA1c had a positive association with GA (r = 0.75, P < 0.01). (3) Using OGTT as golden standard of diabetic diagnosis, receiver operator characteristic (ROC) curve indicated that HbA1c and GA were strong predictors of diabetes. The area under curve (AUC) was 0.882 and 0.881 respectively with no significant difference (P > 0.05). (4) The sensitivity of combined use of HbA1c and GA at optimal cut-off points of 6.1% and 17.1% was significantly higher than that of single use of HbA1c or GA in diabetic screening (94.7% vs 81.1%, 88.4%, P < 0.01). A combined detection of HbA1c and GA may improve the efficacy of diabetic screening. The subject with HbA1c ≥ 6.1% or GA ≥ 17.1% is recommended to undergo OGTT for confirming a diagnosis of diabetes.
    Zhonghua yi xue za zhi 07/2011; 91(26):1813-6.
  • Qing Li, Wu Zhang
    [show abstract] [hide abstract]
    ABSTRACT: In this paper, a Multi-Thread Interface (MTI) is developed into a lightweight and easy-to-use class library. This library is based on multi-core processor by means of packaging Windows thread interface functions, with which serial codes can be expanded into parallel ones. In the library, thread-related operations are enveloped into classes in order that programmers can use them more conveniently. MTI constructs for parallelization, restricting control of threads and communication primitives for synchronization. Using this framework, parallel algorithms have been implemented for computing numerical integral and matrix multiplying to prove that MTI reduces the difficulty, such as "programming wall", brought by multicore system.� �
  • [show abstract] [hide abstract]
    ABSTRACT: With the fast development of the Internet, there are more and more learning resources including various multimedia resources on the Web. Meanwhile, e-learning becomes more popular and important due to its convenience and autonomy. However, how to help users conduct high quality resource retrieval remains a challenge and should be considered first for an e-learning system. The traditional indexing approach which classifies multimedia resources into predefined categories can hardly meet user demands. Collaborative tagging (also known as folksonomy) systems provide users with a simple but powerful mechanism to obtain required multimedia resources. Moreover, feedbacks made by users can be utilized to refine the retrieval results during the interactive process between users and systems. In this paper, we present an interactive tutoring system with personalized multimedia resource search, based on collaborative tagging mechanism. We also describe several application scenarios in which content-based image retrieval (CBIR) is combined with collaborative tagging to facilitate interactive tutoring in our cooking recipe database system.
  • [show abstract] [hide abstract]
    ABSTRACT: As a new application model, cloud computing enables users to distribute virtual resource in cloud data center as demand. However, the performance declines as the increase of users. This paper focuses on the network and I/O performance decline due to the virtual machine image and virtual disk image densely deployment in Cloud computing management platform. The method and model for rapid deployment and resource share are proposed on basis of shared storage and inventory theory which is widely used in commodity market. Experiment shows the method can effectively alleviate the performance decline caused by users' dense image requests in cloud computing management platform.
  • [show abstract] [hide abstract]
    ABSTRACT: As a new technology in IT industry, cloud computing has been much focused by both academia and industry. And many topics in cloud computing are under study. However, as one of the most important issue, billing and pricing has been not so much concerned. In this paper, we propose a novel pay-per-use billing model that can be used for the applications in cloud computing. Built based on leasing instances, the model is put forward according to the analysis of the real applications in the cloud computing environment. That is, with the model, the users will be charged totally by the amount of the resources they have used. Simulation results show that the model is correct and feasible.
    Advances in Grid and Pervasive Computing - 6th International Conference, GPC 2011, Oulu, Finland, May 11-13, 2011. Proceedings; 01/2011
  • Source
    [show abstract] [hide abstract]
    ABSTRACT: Despite the ubiquity of physical obstacles (e.g., buildings, hills, and blindages, etc.) in the real world, most of spatial queries ignore the obstacles. In this article, we study a novel form of continuous nearest-neighbor queries in the presence of obstacles, namely continuous obstructed nearest-neighbor (CONN) search, which considers the impact of obstacles on the distance between objects. Given a data set P, an obstacle set O, and a query line segment q, in a two-dimensional space, a CONN query retrieves the nearest neighbor p &in; P of each point p′ on q according to the obstructed distance, the shortest path between p and p′ without crossing any obstacle in O. We formalize CONN search, analyze its unique properties, and develop algorithms for exact CONN query-processing assuming that both P and O are indexed by conventional data-partitioning indices (e.g., R-trees). Our methods tackle CONN retrieval by performing a single query for the entire query line segment, and only process the data points and obstacles relevant to the final query result via a novel concept of control points and an efficient quadratic-based split point computation approach. Then, we extend our techniques to handle variations of CONN queries, including (1) continuous obstructed k nearest neighbor (COkNN) search which, based on obstructed distances, finds the k (&geq; 1) nearest neighbors (NNs) to every point along q; and (2) trajectory obstructed k nearest-neighbor (TOkNN) search, which, according to obstructed distances, returns the k NNs for each point on an arbitrary trajectory (consisting of several consecutive line segments). Finally, we explore approximate COkNN (ACOkNN) retrieval. Extensive experiments with both real and synthetic datasets demonstrate the efficiency and effectiveness of our proposed algorithms under various experimental settings.
    ACM Trans. Database Syst. 01/2011; 36:9.
  • Source
    [show abstract] [hide abstract]
    ABSTRACT: With the tremendous amount of research publications, recommending relevant papers to researchers to fulfill their information need becomes a significant problem. The major challenge to be tackled by our work is that given a target paper, how to effectively recommend a set of relevant papers from an existing citation network. In this paper, we propose a novel method to address the problem by incorporating various citation relations for a proper set of papers, which are more relevant but with a very limited size. The proposed method has two unique properties. Firstly, a metric called Local Relation Strength is defined to measure the dependency between cited and citing papers. Secondly, a model called Global Relation Strength is proposed to capture the relevance between two papers in the whole citation graph. We evaluate our proposed model on a real-world publication dataset and conduct an extensive comparison with the state-of-the-art baseline methods. The experimental results demonstrate that our method can have a promising improvement over the state-of-the-art techniques.
    Web-Age Information Management - 12th International Conference, WAIM 2011, Wuhan, China, September 14-16, 2011. Proceedings; 01/2011

Publication Stats

914 Citations
62.22 Total Impact Points


  • 2014
    • Lands Department of The Government of the Hong Kong Special Administrative Region
      Hong Kong, Hong Kong
  • 1999–2014
    • City University of Hong Kong
      • Department of Computer Science
      Chiu-lung, Kowloon City, Hong Kong
    • University of Sydney
      Sydney, New South Wales, Australia
  • 2012
    • Soochow University (PRC)
      Wu-hsien, Jiangsu Sheng, China
  • 2010–2012
    • Xin Hua Hospital Affiliated to Shanghai Jiao Tong University School of Medicine
      Shanghai, Shanghai Shi, China
  • 1970–2012
    • The University of Hong Kong
      • • Department of Computer Science
      • • Department of Information Technology & Engineering
      Hong Kong, Hong Kong
  • 2008–2011
    • Shanghai University
      Shanghai, Shanghai Shi, China
    • University of Science and Technology of China
      • School of Computer Science and Technology
      Hefei, Anhui Sheng, China
    • Zhejiang Normal University
      Jinhua, Zhejiang Sheng, China
    • Wuhan University
      Wu-han-shih, Hubei, China
  • 2007–2011
    • USTC-CityU Joint Advanced Research Center
      Hong Kong, Hong Kong
    • Shanghai Clinical Research Center
      Shanghai, Shanghai Shi, China
    • Carnegie Mellon University
      • Computer Science Department
      Pittsburgh, Pennsylvania, United States
  • 2004–2010
    • Shanghai Cancer Institute
      Shanghai, Shanghai Shi, China
  • 2009
    • Zhejiang University
      • College of Computer Science and Technology
      Hangzhou, Zhejiang Sheng, China
  • 2007–2009
    • Arizona State University
      Phoenix, Arizona, United States
  • 2006
    • Renmin University of China
      • School of Information
      Beijing, Beijing Shi, China
  • 1998
    • The Hong Kong Polytechnic University
      • Department of Computing
      Hong Kong, Hong Kong
  • 1997
    • University of New South Wales
      • School of Computer Science and Engineering
      Kensington, New South Wales, Australia
  • 1994–1995
    • The Hong Kong University of Science and Technology
      • Department of Computer Science and Engineering
      Kowloon, Hong Kong
    • University of Melbourne
      Melbourne, Victoria, Australia
  • 1988
    • University of Southern California
      Los Angeles, California, United States