Browsing by Author "Wu, Cheng-Wei"

Now showing 1 - 17 of 17

Analyzing Students' Attention in Class Using Wearable Devices
(2017-01-01)
Zhang, Xin
;
Wu, Cheng-Wei
;
Fournier-Viger, Philippe
;
Van, Lan-Da
;
Tseng, Yu-Chee
Detecting students' attention in class provides key information to teachers to capture and retain students' attention. Traditionally, such information is collected manually by human observers. Wearable devices, which have received a lot of attention recently, are rarely discussed in this field. In view of this, we propose a multimodal system which integrates a head-motion module, a pen-motion module, and a visual-focus module to accurately analyze students' attention levels in class. These modules collect information via cameras, accelerometers, and gyroscopes integrated in wearable devices to recognize students' behaviors. From these behaviors, attention levels are inferred for various time periods using a rule-based approach and a data-driven approach. The former infers a student's attention states using user-defined rules, while the latter relies on hidden relationships in the data. Extensive experimental results show that the proposed system has excellent performance and high accuracy. To the best of our knowledge, this is the first study on attention level inference in class using wearable devices. The outcome of this research has the potential of greatly increasing teaching and learning efficiency in class.
Applying Machine Learning to Head Gesture Recognition using Wearables
(2017-01-01)
Wu, Cheng-Wei
;
Yang, Hua-Zhi
;
Chen, Yan-Ann
;
Ensa, Bajo
;
Ren, Yi
Recently, some studies for head gestures recognition have been proposed, but most of them are based on image processing technology. Moreover, there is less focus on the use of wearable devices in recognizing head gestures. In this paper, we apply machine learning techniques to recognize some common head activities using a head-mounted wearable device. We use the wearable device to collect sensor data related to user's head activities. Then, we apply energy-based segmentation method on the collected data to find out the data segments where the activities may occur. Finally, we extract candidate features from the segments and feed them into a pre-trained classifier to identify the type of head gesture. We implement the prototype of above methods on Arduino platform and evaluate the efficiency of the proposed methods on real datasets. The experiment results show that our proposed methods can effectively and efficiently identify different types of head gestures with an average accuracy rate of 95%.
Efficient Algorithms for Mining Top-K High Utility Itemsets
(2016-01-01)
Tseng, Vincent S.
;
Wu, Cheng-Wei
;
Fournier-Viger, Philippe
;
Yu, Philip S.
;
資訊工程學系
High utility itemsets (HUIs) mining is an emerging topic in data mining, which refers to discovering all itemsets having a utility meeting a user-specified minimum utility threshold min_util. However, setting min_util appropriately is a difficult problem for users. Generally speaking, finding an appropriate minimum utility threshold by trial and error is a tedious process for users. If min_util is set too low, too many HUIs will be generated, which may cause the mining process to be very inefficient. On the other hand, if min_util is set too high, it is likely that no HUIs will be found. In this paper, we address the above issues by proposing a new framework for top-k high utility itemset mining, where k is the desired number of HUIs to be mined. Two types of efficient algorithms named TKU (mining Top-K Utility itemsets) and TKO (mining Top-K utility itemsets in One phase) are proposed for mining such itemsets without the need to set min_util. We provide a structural comparison of the two algorithms with discussions on their advantages and limitations. Empirical evaluations on both real and synthetic datasets show that the performance of the proposed algorithms is close to that of the optimal case of state-of-the-art utility mining algorithms.
Efficient Mining of High-Utility Sequential Rules
(2015-01-01)
Zida, Souleymane
;
Fournier-Viger, Philippe
;
Wu, Cheng-Wei
;
Lin, Jerry Chun-Wei
;
Tseng, Vincent S.
High-utility pattern mining is an important data mining task having wide applications. It consists of discovering patterns generating a high profit in databases. Recently, the task of high-utility sequential pattern mining has emerged to discover patterns generating a high profit in sequences of customer transactions. However, a well-known limitation of sequential patterns is that they do not provide a measure of the confidence or probability that they will be followed. This greatly hampers their usefulness for several real applications such as product recommendation. In this paper, we address this issue by extending the problem of sequential rule mining for utility mining. We propose a novel algorithm named HUSRM (High-Utility Sequential Rule Miner), which includes several optimizations to mine high-utility sequential rules efficiently. An extensive experimental study with four datasets shows that HUSRM is highly efficient and that its optimizations improve its execution time by up to 25 times and its memory usage by up to 50 %.
Efficiently mining high utility sequential patterns in static and streaming data
(2017-01-01)
Zihayat, Morteza
;
Wu, Cheng-Wei
;
An, Aijun
;
Tseng, Vincent S.
;
Lin, Chien
High utility sequential pattern (HUSP) mining has emerged as a novel topic in data mining. Although some preliminary works have been conducted on this topic, they incur the problem of producing a large search space for high utility sequential patterns. In addition, they mainly focus on mining HUSPs in static databases and do not take streaming data into account, where unbounded data come continuously and often at a high speed. To efficiently deal with both problems, we propose a novel framework for mining high utility sequential patterns over static and streaming databases. In this regard, two efficient data structures named ItemUtilLists (Item Utility Lists) and HUSP-Tree (High Utility Sequential Pattern Tree) are proposed to maintain essential information for mining HUSPs in both offline and online fashions. In addition, a novel utility model called Sequence-Suffix Utility is proposed for effectively pruning the search space in HUSP mining. We propose an algorithm named HUSP-Miner (High Utility Sequential Pattern Miner) to find HUSPs in static databases efficiently. Then, a one-pass algorithm named HUSP-Stream (High Utility Sequential Pattern mining over Data Streams) is proposed to incrementally update ItemUtilLists and HUSP-Tree online and find HUSPs over data streams. To the best of our knowledge, HUSP-Stream is the first method to find HUSPs over data streams. Experimental results on both real and synthetic datasets show that HUSP-Miner outperforms the compared algorithms substantially in terms of execution time, memory usage and number of generated candidates. The experiments also demonstrate impressive performance of HUSP-Stream to update the data structures and discover HUSPs over data streams.
EFIM-Closed: Fast and Memory Efficient Discovery of Closed High-Utility Itemsets
(2016)
Fournier-Viger, Philippe
;
Zida, Souleymane
;
Lin, Jerry Chun-Wei
;
Wu, Cheng-Wei
;
Tseng, Vincent S.
Discovering high-utility temsets in transaction databases is a popular data mining task. A limitation of traditional algorithms is that a huge amount of high-utility itemsets may be presented to the user. To provide a concise and lossless representation of results to the user, the concept of closed high-utility itemsets was proposed. However, mining closed high-utility itemsets is computationally expensive. To address this issue, we present a novel algorithm for discovering closed high-utility itemsets, named EFIM-Closed. This algorithm includes novel pruning strategies named closure jumping, forward closure checking and backward closure checking to prune non-closed high-utility itemsets. Furthermore, it also introduces novel utility upper-bounds and a transaction merging mechanism. Experimental results shows that EFIM-Closed can be more than an order of magnitude faster and consumes more than an order of magnitude less memory than the previous state-of-art CHUD algorithm.
EFIM: a fast and memory efficient algorithm for high-utility itemset mining
(2017-05-01)
Zida, Souleymane
;
Fournier-Viger, Philippe
;
Lin, Jerry Chun-Wei
;
Wu, Cheng-Wei
;
Tseng, Vincent S.
In recent years, high-utility itemset mining has emerged as an important data mining task. However, it remains computationally expensive both in terms of runtime and memory consumption. It is thus an important challenge to design more efficient algorithms for this task. In this paper, we address this issue by proposing a novel algorithm named EFIM (EFficient high-utility Itemset Mining), which introduces several new ideas to more efficiently discover high-utility itemsets. EFIM relies on two new upper bounds named revised sub-tree utility and local utility to more effectively prune the search space. It also introduces a novel array-based utility counting technique named Fast Utility Counting to calculate these upper bounds in linear time and space. Moreover, to reduce the cost of database scans, EFIM proposes efficient database projection and transaction merging techniques named High-utility Database Projection and High-utility Transaction Merging (HTM), also performed in linear time. An extensive experimental study on various datasets shows that EFIM is in general two to three orders of magnitude faster than the state-of-art algorithms HUP, HUI-Miner, HUP-Miner, FHM and UP-Growth+ on dense datasets and performs quite well on sparse datasets. Moreover, a key advantage of EFIM is its low memory consumption.
EFIM: A Highly Efficient Algorithm for High-Utility Itemset Mining
(2015-01-01)
Zida, Souleymane
;
Fournier-Viger, Philippe
;
Lin, Jerry Chun-Wei
;
Wu, Cheng-Wei
;
Tseng, Vincent S.
High-utility itemset mining (HUIM) is an important data mining task with wide applications. In this paper, we propose a novel algorithm named EFIM (EFficient high-utility Itemset Mining), which introduces several new ideas to more efficiently discovers high-utility itemsets both in terms of execution time and memory. EFIM relies on two upper-bounds named sub-tree utility and local utility to more effectively prune the search space. It also introduces a novel array-based utility counting technique named Fast Utility Counting to calculate these upper-bounds in linear time and space. Moreover, to reduce the cost of database scans, EFIM proposes efficient database projection and transaction merging techniques. An extensive experimental study on various datasets shows that EFIM is in general two to three orders of magnitude faster and consumes up to eight times less memory than the state-of-art algorithms d 2 HUP, HUI-Miner, HUP-Miner, FHM and UP-Growth+.
Eye On You: Fusing Gesture Data from Depth Camera and Inertial Sensors for Person Identification
(2018-01-01)
Chang, Wei-Chun
;
Wu, Cheng-Wei
;
Tsai, Richard Yi-Chia
;
Lin, Kate Ching-Ju
;
Tseng, Yu-Chee
Person identification (PID) is a key issue in many IoT applications. It has long been studied and achieved by technologies such as RFID and face/fingerprint/iris recognition. These approaches, however, have their limitations due to environmental constraints (such as lighting and obstacles) or require close contact to specific devices. Therefore, their recognition rates highly depend on use scenarios. To enable reliable and remote PID, in this work, we present EOY (Eye On You) 1, a data fusion approach that combines two kinds of sensors, a 3D depth camera and wearable sensors embedded with inertial measurement unit (IMU). Since these two kinds of data share common features, we are able to fuse them to conduct PID. Further, the result can be transferred to a mobile platform (such as robot) since we have less constraints on devices. To realize EOY, we develop fusion algorithms to address practical challenges, such as asynchronous timing and coordinate calibration. The experimental evaluation shows that EOY can achieve the recognition rate of 95% and is very robust even in crowded areas.
Flowtable-Free Routing for Data Center Networks: A Software-Defined Approach
(2017-01-01)
Ren, Yi
;
Tsai, Tsung-Han
;
Huang, Ji-Cheng
;
Wu, Cheng-Wei
;
Tseng, Yu-Chee
The paradigm shift toward SDN has exhibited the following trends: (1) relying on a centralized and more powerful controller to make intelligent decisions, and (2) allowing a set of relatively dumb switches to route packets. Therefore, efficiently looking up the flowtables in forwarding switches to guarantee low latency becomes a critical issue. In this paper, following the similar paradigm, we propose a new routing scheme called KeySet which is flowtable-free and enables constant-time switching at the forwarding switches. Instead of looking up long flowtables, KeySet relies on a residual system to quickly calculate routing paths. A switch only needs to do simple modular arithmetics to obtain a packet's forwarding output port. Moreover, KeySet has a nice fault-tolerant capability because in many cases the controller does not need to update flowtables at switches when a failure occurs. We validate KeySet through extensive simulations by using general as well as Facebook fat-tree topologies. The results show that the KeySet outperforms the KeyFlow scheme [1] by at least 25% in terms of the length of the forwarding label. Moreover, we show that KeySet is very efficient when applied to fat-trees.
iToy: A LEGO-like Solution for Small Scale IoT Applications
(2017-01-01)
Ren, Yi
;
Alfiansyah, Muhammad
;
Wibowo, Nyoto Arif
;
Wu, Cheng-Wei
;
Geng, JieFu
We can find various Internet of Things (IoT) products available in the market to accommodate users needs. Current IoT applications, however, built delicately for their usage with their own sensor devices and Apps. This limitation thus prevents users from handling flexible situations. In this paper, we propose iToy, an LEGO-like solution to integrate various off-the-shelf sensors for different IoT applications. In iToy, the off-the-shelf sensors are like LEGO plastic bricks, which can be assembled and connected to construct IoT applications. Any sensors constructed in an IoT application can be taken apart again and then used to make other IoT applications. We demonstrate through proof-of-concept prototype that iToy is user-friendly and suitable for forming different small scale IoT applications.
Mining Closed(+) High Utility Itemsets without Candidate Generation
(2015)
Wu, Cheng-Wei
;
Fournier-Viger, Philippe
;
Gu, Jia-Yuan
;
Tseng, Vincent S.
;
資訊工程學系
High utility itemsets (HUIs) mining refers to discovering sets of items that not only co-occur but also carry high utilities (e.g., high profits). HUI mining receives extensive attentions in recent years due to the wide applications in various domains like commerce and biomedicine. However, huge number of HUls might be produced to users, which degrades the efficiency of the mining process. A promising solution to this problem is to mine closed(+) high utility itemset (CHUI), a compact and lossless representation of HUls. Nevertheless, existing algorithms incur the problem of producing a large amount of candidates, which degrades the mining performance in terms of time and space. In this paper, a novel algorithm named CHUI-Miner (Closed(+) High Utility ltemset mining without candidates) for mining CHUls is proposed, which directly computes the utility of itemsets without producing candidates. To our best knowledge, this is the first work addressing the issue of mining CHUls without candidate generation. Experimental results show that CHUl-Miner is several orders of magnitude faster than the state-of-the-art algorithms.
Mining High Utility Itemsets in Big Data
(2015-01-01)
Lin, Ying Chun
;
Wu, Cheng-Wei
;
Tseng, Vincent S.
;
資訊工程學系
;
Department of Computer Science
In recent years, extensive studies have been conducted on high utility itemsets (HUI) mining with wide applications. However, most of them assume that data are stored in centralized databases with a single machine performing the mining tasks. Consequently, existing algorithms cannot be applied to the big data environments, where data are often distributed and too large to be dealt with by a single machine. To address this issue, we propose a new framework for mining high utility itemsets in big data. A novel algorithm named PHUI-Growth (Parallel mining High Utility Itemsets by pattern-Growth) is proposed for parallel mining HUIs on Hadoop platform, which inherits several nice properties of Hadoop, including easy deployment, fault recovery, low communication overheads and high scalability. Moreover, it adopts the MapReduce architecture to partition the whole mining tasks into smaller independent subtasks and uses Hadoop distributed file system to manage distributed data so that it allows to parallel discover HUIs from distributed data across multiple commodity computers in a reliable, fault tolerance manner. Experimental results on both synthetic and real datasets show that PHUI-Growth has high performance on large-scale datasets and outperforms state-of-the-art non-parallel type of HUI mining algorithms.
Mining Minimal High-Utility Itemsets
(2016)
Fournier-Viger, Philippe
;
Lin, Jerry Chun-Wei
;
Wu, Cheng-Wei
;
Tseng, Vincent S.
;
Faghihi, Usef
Mining high-utility itemsets (HUIs) is a key data mining task. It consists of discovering groups of items that yield a high profit in transaction databases. A major drawback of traditional high-utility itemset mining algorithms is that they can return a large number of HUIs. Analyzing a large result set can be very time-consuming for users. To address this issue, concise representations of high-utility itemsets have been proposed such as closed HUIs, maximal HUIs and generators of HUIs. In this paper, we explore a novel representation called the minimal high utility itemsets (MinHUIs), defined as the smallest sets of items that generate a high profit, study its properties, and design an efficient algorithm named MinFHM to discover it. An extensive experimental study with real-life datasets shows that mining MinHUIs can be much faster than mining other concise representations or all HUIs, and that it can greatly reduce the size of the result set presented to the user.
Qnalyzer: Queuing Recognition Using Accelerometer and Wi-Fi Signals
(2017-01-01)
Wu, Zone-Ze
;
Wu, Cheng-Wei
;
Van, Lan-Da
;
Tseng, Yu-Chee
;
資訊工程學系
Queuing recognition is a recently new raised research topic, which uses sensors of smartphones to automatically recognize human queuing behaviors. However, existing collaborative approaches need to exchange sensor data among nearby smartphones, causing extra communication overheads and even delay. In view of this, this work proposes a new framework called Qnalyzer for queuing recognition using accelerometer and Wi-Fi signals. It consists of three tiers. The first tier is run by each individual smartphone to identify each user's context without exchanging data with nearby smartphones. A new algorithm called QCF (Queuer and non-queuer ClassiFier) is proposed, which considers mixture features of accelerometer and Wi-Fi signals to effectively identify whether the user is queuing or not. The second tier is an algorithm called QCT (Queuers ClusTering) running at the server side to effectively identify which queuers belong to which queues based on users' movement features. The third tier is an estimation model called QPE (Queue Property Estimation) for measuring waiting time, service time, and queue lengths. The Qnalyzer prototype on Android smartphones and the corresponding performance evaluations under real-life queuing scenarios are implemented. The extensive experiment results show that Qnalyzer achieves good performance with high accuracy.
Time-Dependent Smart Data Pricing Based on Machine Learning
(2017-01-01)
Tsai, Yi-Chia
;
Cheng, Yu-Da
;
Wu, Cheng-Wei
;
Lai, Yueh-Ting
;
Hu, Wan-Hsun
The purpose of time-dependent smart data pricing (abbreviated as TDP) is to relieve network congestion by offering network users different prices over varied periods. However, traditional TDP has not considered applying machine learning concepts in determining prices. In this paper, we propose a new framework for TDP based on machine learning concepts. We propose two different pricing algorithms, named TDP-TR (TDP based on Transition Rules) and TDP-KNN (TDP based on K-Nearest Neighbors). TDP-TR determines prices based on users' past willingness to pay given different prices, while TDP-KNN determines prices based on the similarity of users' past network usages. The main merit of TDP-TR is low computational cost, while that of TDP-KNN is low maintenance cost. Experimental results on simulated datasets show that the proposed algorithms have good performance and profitability.
UP-Miner: A Utility Pattern Mining Toolbox
(2015)
Tseng, Vincent S.
;
Wu, Cheng-Wei
;
Lin, Jun-Han
;
Fournier-Viger, Philippe
;
資訊工程學系
Utility pattern mining has received extensive attentions in recent years due to the wide and novel applications in various fields like e-commerce, Web mining, finance, biomedicine, etc. However, there exists not yet a toolbox for utility pattern mining so far. In this work, we address this issue by proposing a first-of-its-kind toolbox named UP-Miner (Utility Pattern Miner) that provides various functions for utility pattern mining. The main merits of UP-Miner lie in three aspects: First, it offers implementations of thirteen state-of-the-art algorithms for efficiently mining different types of utility patterns, such as high utility itemsets, high utility episodes and utility-based sequential patterns, as well as four functionalities for processing utility-based databases. Second, it is a cross-platform system implemented in Java with a user-friendly graphical interface. Third, the toolbox and relevant materials, including source codes, benchmark datasets and data generators, are made public on Web (http://bigdatalab.cs.nctu.edu.tw/software.htm) for benefiting the research community.