Kun-Ming Yu

Load balancing approach parallel algorithm for frequent pattern mining

Association rules mining from transaction-oriented databases is an important issue in data mining. Frequent pattern is crucial for association rules generation, time series analysis, classification, etc. There are two categories of algorithms that had been proposed, candidate set generate-and-test approach (Apriori-like) and Pattern growth approach. Many methods had been proposed to solve the association rules mining problem based on FP-tree instead of Apriori-like, since apriori-like algorithm scans the database many times. However, the computation time is costly when the database size is large with FP-tree data structure. Parallel and distributed computing is a good strategy to solve this circumstance. Some parallel algorithms had been proposed, however, most of them did not consider the load balancing issue. In this paper, we proposed a parallel and distributed mining algorithm based on FP-tree structure, Load Balancing FP-Tree (LFP-tree). The algorithm divides the item set for mining by evaluating the trees width and depth. Moreover, a simple and trusty calculate formulation for loading degree is proposed. The experimental results show that LFP-tree can reduce the computation time and has less idle time compared with Parallel FP-Tree (PFP-tree). In addition, it has better speed-up ratio than PFP-tree when number of processors grow. The communication time can be reduced by preserving the heavy loading items in their local computing node.

Tidset-based parallel FP-tree algorithm for the frequent pattern mining problem on PC clusters

Mining association rules from a transaction-oriented database is a problem in data mining. Frequent patterns are essential for generating association rules, time series analysis, classification, etc. There are two categories of algorithms for data mining, the generate-and-test approach (Apriori-like) and the pattern growth approach (FP-tree). Recently, many methods have been proposed for solving this problem based on an FP-tree as a replacement for Apriori-like algorithms, because these need to scan the database many times. However, even for the pattern growth method, the execution time takes long when the database is large or the given support is low. Parallel- distributed computing is good strategy for solving this problem. Some parallel algorithms have been proposed, however, the execution time increases rapidly when the database increases or when the given minimum threshold is small. In this study, an efficient parallel- distributed mining algorithm based on an FP-tree structure - the Tidset-based Parallel FP-tree (TPFP-tree) - is proposed. In order to exchange transactions efficiently, transaction identification set (Tidset) was used to directly choose transactions without scanning databases. The algorithm was verified on a Linux cluster with 16 computing nodes. It was also compared with a PFP-tree algorithm. The dataset generated by IBMs Quest Synthetic Data Generator to verify the performance of algorithms was used. The experimental results showed that this algorithm can reduce the execution time when the database grows. Moreover, it was also observed that this algorithm had better scalability than the PFP-tree.

A Density-Based Algorithm for Redundant Reader Elimination in a RFID Network

Radio frequency identification (RFID) technology has become more sophisticated in recent years and is being developed rapidly for a variety of applications. The problem of eliminating redundant readers has been reduced to the minimum cover problem and proved to be NP-hard. In this paper, a density-based redundant reader elimination algorithm (DRRE) is presented; DRRE eliminates the redundant readers without influencing the number of usable tags. Simulation results demonstrate that the DRRE algorithm performed better in all instances in a variety of environments compared with other algorithms. In a densely deployed RFID network, the DRRE algorithm detected 85% more redundant readers than others.

A weighted load-balancing parallel Apriori algorithm for association rule mining

Because of the exponential growth in worldwide information, companies have to deal with an ever growing amount of digital information. One of the most important challenges for data mining is quickly and correctly finding the relationship between data. The Apriori algorithm is the most popular technique in association rules mining; however, when applying this method, a database has to be scanned many times and many candidate itemsets are generated. Parallel computing is an effective strategy for accelerating the mining process. In this paper, the weighted distributed parallel apriori algorithm (WDPA) is presented as a solution to this problem. In the proposed method, metadata are stored in TID forms, thus only a single scan to the database is needed. The TID counts are also taken into consideration, and therefore better load-balancing as well as reducing idle time for processors can be achieved. According to the experimental results, WDPA outperforms other algorithms while having lower minimum support.

An Ad Hoc Routing Protocol with Multiple Backup Routes

Numerous routing protocols have recently been developed for ad hoc mobile networks. Routing protocols to date can be categorized as either ‘table-driven’ or ‘on-demand’. Many of the proposed routing protocols take the on-demand approach because this does not require keeping lots of routing information. However, these kinds of protocols are not able to react fast enough to maintain routing. In this paper, we propose a new protocol to improve existing on-demand routing protocols by constructing multiple backup routes; when the network topology changed, the proposed protocol could transmit data packets dynamically through backup routes. We then developed an analytic model to estimate the reconnection probability of the proposed algorithm. We also examined the performance by simulating the protocol using ns2. The experimental results showed that the protocol had fewer control packages, lower routing packet overhead, and a higher receiving ratio than others.

Design and Implementation of a RFID Based Real-Time Location-Aware System in Clean Room

Application of Automated Material Handling System (AMHS) to the semiconductor manufacturing factories is one of the most important steps in the automation process of semiconductor business. All cassettes in the vehicle of the transportation facility can send their positioning information through this system. Nevertheless, some positioning information may still be interfered or get lost if there is any relocation of the cassette inside the effective ranges of AMHS by operators. In order to retrieve these lost cassettes only manpower is possible, and the relocation process is time wasting and inefficient. In this paper, we have implemented a Clean Room Real-Time Location-aware System (CRRLS) that is based on the technology of RFID (Radio Frequency Identification). The proposed system can reveal the positioning information of the cassette in the clean room by analyzing the received radio frequency energy information. Moreover, we establish a transmitting model of radio signal, and used this model to estimate distance between RFID tag and readers. A triangular positioning algorithm is developed to estimate the location and its 2-dimensional coordinate of the cassette. Furthermore, reusable RFID tags can save additional information sent by the transceivers and data of parameters for the production process to reduce the possibility of severe damage because of wrong parameters setting. As a result, competitive advantage of the semiconductor business can herewith greatly increase.

Efficient Scheduling Algorithms for Bluetooth Scatternets

When more than seven devices are connected in a Bluetooth scatternet, bridge devices are used to connect two piconets to the scatternet. To deal with possible data transmissions between different piconets, the bridge device must frequently switch to different masters. Suppose, however, that a bridge is serving a piconet and the master in another piconet is calling it at the same time, the calling master has to wait until the bridge completes the previous service. Such transmission delay may accumulate over a long period and the performance of the whole Bluetooth network will degrade significantly. In this work, two new scheduling protocols, namely the static schedule and the hybrid schedule were implemented in an effort to smooth this kind of transmission delay in Bluetooth networks. In this static schedule the rendezvous points between piconets are coordinated by distributing them by using a graph edge coloring technique. In case of a heavy traffic load, the static schedule is expected to perform well. On the other hand, in case of a light traffic load, the static schedule may cause long and unavoidable routing delays even when there is no transmission between piconets; in this case a naive random round-robin (RR) schedule in each piconet is more appropriate. Thus, in the hybrid schedule, each master initially runs a RR scheme in its piconet. When the traffic load is heavier than a predefined threshold value, it runs the static schedule. Finally, simulations were conducted by using an ns-2 simulator and Bluehoc to demonstrate the efficiency and effectiveness of the proposed scheduling protocols.

Accelerating Parallel Frequent Itemset Mining on Graphics Processors with Sorting

Frequent Itemset Mining FIM is one of the most investigated fields of data mining. The goal of Frequent Itemset Mining FIM is to find the most frequently-occurring subsets from the transactions within a database. Many methods have been proposed to solve this problem, and the Apriori algorithm is one of the best known methods for frequent Itemset mining FIM in a transactional database. In this paper, a parallel Frequent Itemset Mining Algorithm, called Accelerating Parallel Frequent Itemset Mining on Graphic Processors with Sorting APFMS, is presented. This algorithm utilizes new-generation graphic processing units GPUs to accelerate the mining process. In it, massive processing units of GPU were used to speed up the frequent item verification procedure on the OpenCL platform. The experimental results demonstrated that the proposed algorithm had dramatically reduced computation time compared with previous methods.

An OpenCL Candidate Slicing Frequent Pattern Mining algorithm on graphic processing units

Frequent pattern mining (FPM) is important in data mining field with Apriori algorithm to be one of the commonly used approaches to solve it. However, Aprori algorithm encounters an issue that the computation time increases dramatically when data size increases and when the threshold is small. Many parallel algorithms have been proposed to speed up the computation using computer clusters or grid systems. GPUs have also been applied on FPM with only few adopting OpenCL although OpenCL has the advantage of being platform independent. Thus, the aim of this research is to develop efficient parallel Aprori strategy using GPU and OpenCL. Our novel method, Candidate Slicing Frequent Pattern Mining (CSFPM) algorithm, improves over the previous method by slicing candidate information to better balance the load between processing units. This strategy is proved to be more efficient according to our experiments. For example, CSFPM is at most 2.6 times faster than the previous method. Therefore, CSFPM is an efficient parallel Apriori algorithm which can reduce computation time and improve overall performance.

The Design and Implementation of a Mobile Location-Aware Digital Signage System

The worldwide Digital Signage market has been getting increasingly popular in recent years. Nevertheless, for service providers, the Digital Signage business is still not easy to manage and time-consuming to operate. In this paper, based on the GPS and wireless infrastructure, we present a mobile location-aware digital signage system (LDSS). Through the centralized system architecture, the applications for the Digital Signage can be expanded to various advertising vehicles including buses and other mobile advertising vehicles. Also, we present an advertisement recommendation algorithm, by which the advertisement can be broadcasted in the right markets more effectively. Evidenced by simulated experiments and analysis, compared with traditional region trigger advertising, sequential advertising and random advertising, our advertisement recommendation algorithm can effectively reach to the key target audiences in planned regions, while show excellent ability to differentiate markets in unplanned regions reaching to the target consumer groups as well.