Wonik Choi

Multi-criteria decision making with skyline computation

Multi-criteria decision making is one of the most critical and yet most challenging components in modern enterprise business intelligence. It is well known that complex business decisions are often made based on multi-dimensional criteria. The competitiveness of optimal business decision making typically resorts to finding a good trade-off among many different and possibly contradicting criteria, e.g., maximum profit, minimum price, minimum resource consumption. A skyline query operator is by design to find the set of interesting data points (objects) over a large dimensional data collection, satisfying a set of possibly contradicting conditions. In this paper, we provide an in-depth coverage of skyline computation models, algorithms and optimization techniques for improving both efficiency and quality of multi-criteria decision making. By reviewing and revising the state of art research in multi-criteria decision making using skyline operations, we describe the essential concepts, the alternative models, and the suite of techniques for providing scalable and elastic skyline computation in massively distributed computing environments. The paper consists of four parts. First, we provide an overview of skyline query operators in terms of concepts, basic processing algorithms and representative application scenarios. Second, we review the state of art literature in skyline query processing research and development, outline the most representative classes of skyline query processing and optimization techniques and discuss the pros and cons of existing approaches. Third, we provide a comprehensive analysis on the inherent limitations of some existing skyline models and algorithms and discuss why scaling skyline query processing over large high dimensional datasets continues to pose daunting challenges. Finally, we present optimization techniques for designing parallel skyline query processing algorithms and how to utilize GPUs to support and scale parallel skyline computations over high dimensional large datasets. We also introduce a novel dominance test technique, called GNL (GPU-based Nested Loop), which can drastically reduce the cost of dominance tests by leveraging GPUs, and outperform CPU-based dominance tests.

Adaptive cell-based index for moving objects

R-tree based access methods for moving objects are hardly applicable in practice, due mainly to excessive space requirements and high management costs. To overcome the limitations of such R-tree based access methods, we propose a new index structure called AIM (Adaptive cell-based Index for Moving objects). The AIM is a cell-based multiversion access structure adopting an overlapping technique. The AIM refines cells adaptively to handle regional data skew, which may change its locations over time. Through the extensive performance studies, we observed that The AIM consumed at most 30% of the space required by R-tree based methods, and achieved higher query performance compared with R-tree based methods.

An adaptive hashing technique for indexing moving objects

Although hashing techniques are widely used for indexing moving objects, they cannot handle the dynamic workload, e.g. the traffic at peak hour vs. that in the night. This paper proposes an adaptive hashing technique to support the dynamic workload efficiently. The proposed technique maintains two levels of the hashes, one for fast moving objects and the other for quasi-static objects. A moving object changes its level adaptively according to the degree of its movement. We also present the theoretical analysis and experimental results which show that the proposed approach is more suitable than the basic hashing under the dynamic workload.

Spatio-temporal data warehouses using an adaptive cell-based approach

Most of the framework for supporting OLAP operations over immense amounts of spatio-temporal data is based on multi-tree structures. The multi-tree frameworks, however, are hardly applicable to spatio-temporal OLAP in practice, due mainly to high management costs and low query efficiency. To overcome the limitations of such multi-tree frameworks, we propose a new approach called ST-Cube (spatio-temporal cube), which is an adaptive cell-based, total-ordered and prefix-summed cube for spatio-temporal data warehouses. Our extensive performance studies show that the ST-Cube requires less space and achieves higher query performance than multi-tree frameworks, under various operational conditions.

Distance Based Pre-cluster Head Selection Scheme for a Chain-Based Protocol

PEGASIS, a chain-based protocol, forms chains from sensor nodes so that each node transmits and receives from a neighbor. In this way, only one node (known as a HEAD) is selected from that chain to transmit to the sink. Although PEGASIS is able to balance the workload among all of the nodes by selecting the HEAD node in turn, a considerable amount of energy may be wasted when nodes which are far away from sink node act as the HEAD. In this study, DERP (Distance-based Energy-efficient Routing Protocol) is proposed to address this problem. DERP is a chain-based protocol that improves the greedy-algorithm in PEGASIS by taking into account the distance from the HEAD to the sink node. The main idea of DERP is to adopt a pre-HEAD (P-HD) to distribute the energy load evenly among sensor nodes. In addition, to scale DERP to a large network, it can be extended to a multi-hop clustering protocol by selecting a “relay node” according to the distance between the P-HD and SINK. Analysis and simulation studies of DERP show that it consumes up to 80% less energy, and has less of a transmission delay compared to PEGASIS.

Similarity search using the polar wavelet in time series databases

In this paper, we propose the novel feature extraction method, called the Polar wavelet, which can improve the search performance for locally distributed time series data. Among various feature extraction methods, the Harr wavelet has been popularly used to extract features from time series data. However, theHarr wavelet does not show the good performance for sequences of similar averages. The proposed method uses polar coordinates which are not affected by averages and can reduce the search space efficiently without false dismissals. The experiments are performed on real temperature dataset to verify the performance of the proposed method.

Adaptive Mobile Sink Path Based Energy Efficient Routing Protocol for Wireless Sensor Network

In this paper, we propose a novel approach to optimize the movement of mobile sink node, called AMSP(Adaptive Mobile Sink Path) for mobile sensor network environments. Currently available studies usually suffer from unnecessary data transmission resulting from random way point approach. To address the problem, we propose a method which uses the Hilbert curve to create a path. The proposed method guarantees shorten transmission distance between the sink node and each sensor node by assigning orders of the curve according to sensor node density. Furthermore, The schedule of the sink node is informed to all of the sensing nodes so that the Duty Cycle helps the network be more energy efficient. In our experiments, the proposed method outperforms the existing works such as TTDD and CBPER by up to 80% in energy consumption.

Lifetime Prediction Routing Protocol for Wireless Sensor Networks

We propose LPDD (Lifetime Prediction Directed Diffusion), a novel energy-aware routing protocol for sensor networks that aims at increasing network survivability without a significant increase in latency. The key concept behind the protocol is the adaptive selection of routes by predicting the battery lifetime of the minimum energy nodes along the routes.

Generating a New Dataset for Korean Scene Text Recognition with Augmentation Techniques

Korean text recognition in a natural scene is a challenging task due to the complexity of character shapes and the lack of dataset comparing to English or other languages. In this paper, we present a new dataset with the goal of improving the recognition of Korean natural scene text. Our dataset is generated by data augmentation techniques without losing a reality. The number of augmented images is 3 million and these images are made up of about 30 non-commercial fonts and 511,000 words from a standard Korean language dictionary. This enormous amount of data offers new possibilities for training deeper neural networks. In our extensive experiments, results show that our dataset effectively trains convolutional recurrent neural networks that achieve state-of-the-art performance on the Korea Advanced Institute of Science & Technology (KAIST) scene text database with very few data-acquisition costs.

Cloud-Based Positioning Method with Visualized Signal Images

This paper presents a cloud-based positioning method that leverages visualized signal images through visual analytics and deep learning. At a mobile client, such as a smart phone, this approach transforms multidimensional signals captured at a known location and at a given time into a signal image and transmits such visual signal images to the Cloud. By collecting and storing many such visual signal images as fingerprints, we can build a visual signal image cloud and produce a signal image map for the geographical region of interest and utilize such signal image map to serve the positioning requests of mobile clients on the move. When a user Alice wants to know her current position, her mobile client will generate a signal image from the multiple signals it receives with timestamp and send this query image to a Cloud server. The server searches the existing signal images stored in the cloud to find those that are similar to this query signal image and estimates the positioning of Alice based on the locations of those similar signal images collected by the server. We evaluate our visual signal images based positioning system on the entire two floors of a large department store and on the street and shops outside the department store, with the signal images collected over 30 minutes before serving positioning queries. The mean error of up to 4 meters is observed. To further verify the applicability of the proposed method, extensive experiments were conducted to distinguish whether a user is indoor or outdoor by applying a deep learning algorithm with 60% of signal images collected for training and 40% signal images for testing. This experiment shows that the proposed method is able to distinguish indoor and outdoor with accuracy of about 95%.