Browsing by Author "Cao, Dongwei"
Now showing 1 - 3 of 3
Results Per Page
Sort Options
Item Delineation of the Stiff Layer from FWD Measurements(Minnesota Department of Transportation, 2001-10) Guzina, Bojan; Cao, DongweiThe Falling Weight Deflectometer (FWD) is a widely used non-destructive test device for estimating the pavement stiffness properties. However, the conventional elastostatic interpretation of FWD measurements is generally associated with a number of inconsistencies. The purpose of this project is to develop a reliable and effective dynamic backcalculation method capable of estimating the location and properties of the permanent or seasonal stiff layer (as well as other pavement stiffness properties) from FWD measurements. The backcalculation method is implemented in the form of a user-friendly software that allows unedited deflection time histories from the FWD test to be used as an input to the back-analysis. The backcalculation scheme developed in this study is based on the Artificial Neural Network (ANN) approach and employs a three-dimensional multilayer viscoelastic dynamic model as a predictive tool.Item Recognizing and Learning Unknown Emerging Concepts(2004-05-24) Cao, Dongwei; Boley, DanielWe study the classification of data in which some of the concepts represented by the data are known in advance, while new, emerging, concepts are discovered as they appear in the data. The resulting paradigm is called Concept Emergence. Unlike clustering, we start with some known classes, but then learn emerging concepts using new unlabeled data. Unlike Concept Drift, we assume the original concepts remain stationary. This differs from outlier detection because we use the rejected samples to update the classifier. We illustrate the method on both synthetic and real data sets using Support Vector Machine classifiers.Item Training Support Vector Machine using Adaptive Clustering(2003-10-07) Cao, Dongwei; Boley, DanielTraining support vector machines involves a huge optimization problem and many specially designed algorithms have been proposed. In this paper, we proposed an algorithm called ClusterSVM that accelerates the training process by exploiting the distributional properties of the training data, that is, the natural clustering of the training data and the overall layout of these clusters relative to the decision boundary of support vector machines. The proposed algorithm first partitions the training data into several pair-wise disjoint clusters. Then, the representatives of these clusters are used to train an initial support vector machine, based on which we can approximately identify the support vectors and non-support vectors. After replacing the cluster containing only non-support vectors with its representative, the number of training data can be significantly reduced, thereby speeding up the training process. The proposed ClusterSVM has been tested against the popular training algorithm SMO on both the artificial data and the real data, and a significant speedup was observed. The complexity of ClusterSVM scales with the square of the number of support vectors and, after a further improvement, it is expected that it will scale with square of the number of non-boundary support vectors.