Browsing by Author "Liu, Chien-Liang"

Now showing 1 - 20 of 37

A fall detection system using k-nearest neighbor classifier
(2010-10-01)
Liu, Chien-Liang
;
Lee, Chia-Hoang
;
Lin, Ping-Min
;
資訊工程學系
;
Department of Computer Science
The main purpose of this paper is to use off-the-shelf devices to develop a fall detection system. In human body identification, human body silhouette is adopted to improve privacy protection, and vertical projection histograms of the silhouette image and statistical scheme are used to reduce the effect of human body upper limb activities. The kNN classification algorithm is used to classify the postures using the ratio and difference of human body silhouette bounding box height and width. Meanwhile, since time difference is a vital factor to differentiate fall incident event and lying down event, the critical time difference is obtained from the experiment and verified by statistical hypothesis testing. With the help of the kNN classifier and the critical time difference, a fall incident detection system is developed to detect fall incident events. The experiment shows that it could reduce the effect of upper limb activities and the system has a correct rate of 84.44% on fall detection and lying down event detection. (C) 2010 Elsevier Ltd. All rights reserved.
A fall detection system using k-nearest neighbor classifier
(2010-10-01)
Liu, Chien-Liang
;
Lee, Chia-Hoang
;
Lin, Ping-Min
;
資訊工程學系
;
Department of Computer Science
The main purpose of this paper is to use off-the-shelf devices to develop a fall detection system. In human body identification, human body silhouette is adopted to improve privacy protection, and vertical projection histograms of the silhouette image and statistical scheme are used to reduce the effect of human body upper limb activities. The kNN classification algorithm is used to classify the postures using the ratio and difference of human body silhouette bounding box height and width. Meanwhile, since time difference is a vital factor to differentiate fall incident event and lying down event, the critical time difference is obtained from the experiment and verified by statistical hypothesis testing. With the help of the kNN classifier and the critical time difference, a fall incident detection system is developed to detect fall incident events. The experiment shows that it could reduce the effect of upper limb activities and the system has a correct rate of 84.44% on fall detection and lying down event detection. (C) 2010 Elsevier Ltd. All rights reserved.
A predictive model for acute allograft rejection of liver transplantation
(2018-03-15)
Liu, Chien-Liang
;
Soong, Ruey-Shyang
;
Lee, Wei-Chen
;
Chen, De-Hsuan
;
Hsu, Shang-Hwa
Orthotopic liver transplantation (OLT) has become an increasingly used treatment for end-stage liver disease. However, acute allograft rejection is still a problem in postoperative care of liver transplantation with immunosuppressive therapy and it can lead to allograft damage and harm the survival of liver transplantation patient. This work proposes to use data-driven approach to build a predictive model for acute rejection. We consider not only prediction accuracy, but also interpretability of the prediction outcome in building the predictive model, so that the medical staffs can identify how the prediction is induced from data. The experiments use the real data provided by liver transplantation intensive care unit (ICU) of Chang Gung Memorial Hospital, Taiwan. In this work, the data is from a medical center, in which the patient data ranges from 2004 to 2013, and the number of data records is approximately 2 million. To the best of our knowledge, this is the first work using a large-scale database to focus on liver transplantation and generate interpretable rules that could be used by medical staffs. We compare with several methods, including SVM, ANN and random forest, and the experimental results indicate that the proposed method is comparative, and provides interpretable results. Central to the proposed method is to consider interpretability, and the goal is to provide interpretable results for the medical staffs to make decisions. The proposed transformation algorithms belong to data-driven approaches, so they could be applied to other intelligent or expert systems. Moreover, the outcomes are presented in rule format, which could be used by medical staffs and other expert systems. (C) 2017 Elsevier Ltd. All rights reserved.
Actor-Critic Deep Reinforcement Learning for Solving Job Shop Scheduling Problems
(2020-01-01)
Liu, Chien-Liang
;
Chang, Chuan-Chin
;
Tseng, Chun-Jan
;
工業工程與管理學系
;
Department of Industrial Engineering and Management
In the past decades, many optimization methods have been devised and applied to job shop scheduling problem (JSSP) to find the optimal solution. Many methods assumed that the scheduling results were applied to static environments, but the whole environments in the real world are always dynamic. Moreover, many unexpected events such as machine breakdowns and material problems may be present to adversely affect the initial job scheduling. This work views JSSP as a sequential decision making problem and proposes to use deep reinforcement learning to cope with this problem. The combination of deep learning and reinforcement learning avoids handcraft features as used in traditional reinforcement learning, and it is expected that the combination will make the whole learning phase more efficient. Our proposed model comprises actor network and critic network, both including convolution layers and fully connected layer. Actor network agent learns how to behave in different situations, while critic network helps agent evaluate the value of statement then return to actor network. This work proposes a parallel training method, combining asynchronous update as well as deep deterministic policy gradient (DDPG), to train the model. The whole network is trained with parallel training on a multi-agent environment and different simple dispatching rules are considered as actions. We evaluate our proposed model on more than ten instances that are present in a famous benchmark problem library & x2013; OR library. The evaluation results indicate that our method is comparative in static JSSP benchmark problems, and achieves a good balance between makespan and execution time in dynamic environments. Scheduling score of our method is 91.12 & x0025; in static JSSP benchmark problems, and 80.78 & x0025; in dynamic environments.
An HMM-Based Algorithm for Content Ranking and Coherence-Feature Extraction
(2013-03-01)
Liu, Chien-Liang
;
Hsaio, Wen-Hoar
;
Lee, Chia-Hoang
;
Chi, Hsiao-Cheng
;
資訊工程學系
In this paper, we propose an algorithm called coherence hidden Markov model (HMM) to extract coherence features and rank content. Coherence HMM is a variant of HMM and is used to model the stochastic process of essay writing and identify topics as hidden states, given sequenced clauses as observations. This study uses probabilistic latent semantic analysis for parameter estimation of coherence HMM. In coherence-feature extraction, support vector regression (SVR) with surface features and coherence features is used for essay grading. The experimental results indicate that SVR can benefit from coherence features. The adjacent agreement rate and the exact agreement rate are 95.24% and 59.80%, respectively. Moreover, this study submits high-scoring essays to the same experiment and finds that the adjacent agreement rate and exact agreement rate are 98.33% and 64.50%, respectively. In content ranking, we design and implement an intelligent assisted blog writing system based on the coherence-HMM ranking model. Several corpora are employed to help users efficiently compose blog articles. When users finish composing a clause or sentence, the system provides candidate texts for their reference based on current clause or sentence content. The experimental results demonstrate that all participants can benefit from the system and save considerable time on writing articles.
An Unsupervised Automated Essay-Scoring System
(2010-09-01)
Chen, Yen-Yu
;
Liu, Chien-Liang
;
Lee, Chia-Hoang
;
Chang, Tao-Hsing
;
資訊工程學系
Background music recommendation based on latent factors and moods
(2018-11-01)
Liu, Chien-Liang
;
Chen, Ying-Chuan
;
資訊工程學系
;
工業工程與管理學系
;
Department of Computer Science
Many mobile devices are equipped with video shooting function, and users tend to use these mobile devices to produce user generated content (UGC), and share with friends or the public owing to the popularity of social media. To make the video to be attractive, embedding appropriate background music into the video is a popular way to enrich user experience, but it is a time-consuming and labor-intensive task to find music that fits the video. This work proposes to use latent factors to recommend a list of music songs for a given video, in which the recommendation is based on the proposed score function, which involves the weighted average of the latent factors for the video and music. Moreover, we use pairwise ranking to design the objective function, and use stochastic gradient descent to optimize the proposed objective function. In the experiments, we specify two hypotheses and design several experiments to assess the performance and the effectiveness of the proposed algorithm from different aspects, including accuracy, quantitative research, and qualitative research. The experimental results indicate that the proposed model is promising in accuracy and quantitative research. Furthermore, this work provides detailed analysis to investigate the fitness of the background music that recommended by the system through interviewing the subjects.
Bayesian exploratory clustering with entropy Chinese restaurant process
(2018-01-01)
Liu, Chien-Liang
;
Hsaio, Wen-Hoar
;
Lin, Che-Yuan
;
資訊工程學系
;
工業工程與管理學系
Data exploration is essential to data analytics, especially when one is confronted with massive datasets. Clustering is a commonly used technique in data exploration, since it can automatically group data instances into a list of meaningful categories, and capture the natural structure of data. Traditional finite mixture model requires the number of clusters to be specified in advance of analyzing the data, and this parameter is crucial to the clustering performance. Chinese restaurant process (CRP) mixture model provides an alternative to this problem, allowing the model complexity to grow as more data instances are observed. Although CRP provides the flexibility to create a new cluster for subsequent data instances, one still has to determine the hyperparameter of the prior and the parameters for the base distribution in the likelihood part. This work proposes a non-parametric clustering algorithm based on CRP with two main differences. First, we propose to create a new cluster based on entropy of the posterior, whereas the CRP uses a hyperparameter to control the probability of creating a new cluster. Second, we propose to dynamically adjust the parameters of the base distribution according to the mean of the observed data owing to Chebyshev's inequality. Additionally, detailed derivation and update rules are provided to perform posterior inference with the proposed collapsed Gibbs sampling algorithm. The experimental results indicate that the proposed algorithm avoids to specify the number of clusters and works well on several datasets.
Clustering data with partial background information
(2019-05-01)
Liu, Chien-Liang
;
Hsaio, Wen-Hoar
;
Chang, Tao-Hsing
;
Li, Hsuan-Hsun
;
資訊工程學系
Clustering with partial supervision background information or semi-supervised clustering, learning from a combination of both labeled and unlabeled data, has received a lot of attention over the last decade. The supervisory information is usually used as the constraints to bias clustering towards a good region of search space. This paper proposes a semi-supervised algorithm, called constrained non-negative matrix factorization (Constrained-NMF), with a few labeled examples as constraints to improve performance. The proposed algorithm is a matrix factorization algorithm, in which initialization of matrices is required at the beginning. Although the benefits of good initialization are well-known, randomized seeding of basis matrix and coefficient matrix is still the standard approach for many non-negative matrix factorization (NMF) algorithms. This work devises an algorithm called entropy-based weighted semi-supervised fuzzy c-means (EWSS-FCM) algorithm to initialize the seeds. The experimental results indicate that the proposed Constrained-NMF can benefit from the initialization obtained from EWSS-FCM, which emphasizes the role of labeled examples and automatically weights them during the course of clustering. This work considers labeled examples in the objective functions to devise the two algorithms, in which the labeled information is propagated to unlabeled examples iteratively. We further analyze the proposed Constrained-NMF and give convergence justifications. The experiments are conducted on five real data sets, and experimental results indicate that the proposed algorithm generally outperforms the other alternatives.
Clustering documents with labeled and unlabeled documents using fuzzy semi-Kmeans
(2013-06-16)
Liu, Chien-Liang
;
Chang, Tao-Hsing
;
Li, Hsuan-Hsun
;
資訊工程學系
;
Department of Computer Science
While focusing on document clustering, this work presents a fuzzy semi-supervised clustering algorithm called fuzzy semi-Kmeans. The fuzzy semi-Kmeans is an extension of K-means clustering model, and it is inspired by an EM algorithm and a Gaussian mixture model. Additionally, the fuzzy semi-Kmeans provides the flexibility to employ different fuzzy membership functions to measure the distance between data. This work employs Gaussian weighting function to conduct experiments, but cosine similarity function can be used as well. This work conducts experiments on three data sets and compares fuzzy semi-Kmeans with several methods. The experimental results indicate that fuzzy semi-Kmeans can generally outperform the other methods. (C) 2013 Elsevier B.V. All rights reserved.
Database-driven Framework for RDF Statements Generation
(2011)
Liu, Chien-Liang
;
Lee, Chia-Hoang
;
資訊工程學系
;
Department of Computer Science
The web has become the biggest database in the world. However, most web content today is designed for humans to read, not for computer programs to manipulate meaningfully. The first step to realize semantic web is putting data on the Web in a form that machines can naturally understand. The main objective of the work reported in this paper is to provide a database-driven framework to automatically extract relational database schema and transform the data into RDF statements representation. Moreover, we employ the Wikipedia to obtain meta-information about the concept. Then, the NGD score is used to measure the relatedness of the terms and the hyponym patterns are further used to discover "is_a" relationship. As our experience with a first simple prototype has shown that this approach could be applied to the environment where the web page content is stored in the database.
Deep 3D Convolutional Neural Network Architectures for Alzheimer's Disease Diagnosis
(2018-01-01)
Karasawa, Hiroki
;
Liu, Chien-Liang
;
Ohwada, Hayato
;
交大名義發表
;
National Chiao Tung University
Dementia has become a social problem in the aging society of advanced countries. Currently, 46.8 million people have dementia worldwide, and that figure is predicted to increase threefold to 130 million people by 2050. Alzheimer's disease (AD) is the most common form of dementia. The cost of care for AD patients in 2015 was 818 billion US dollars and is expected to increase dramatically in the future, due to the increasing number of patients as a result of the aging society. However, it is still very difficult to cure AD; thus, the detection of AD is crucial. This study proposes the use of machine learning to detect AD using brain image data, with the goal of reducing the cost of diagnosing and caring for AD patients. Most machine learning algorithms rely on good feature representations, which are commonly obtained manually and require domain experts to provide guidance. Feature extraction is a time-consuming and labor-intensive task. In contrast, the 3D Convolutional Neural Network (3DCNN) automatically learns feature representation from images and is not greatly affected by image processing. However, the performance of CNN depends on its layer architecture. This study proposes a novel 3DCNN architecture for MRI image diagnosis of AD.
Deep Discriminative Features Learning and Sampling for Imbalanced Data Problem
(2018-01-01)
Liu, Yi-Hsun
;
Liu, Chien-Liang
;
Tseng, Vincent Shin-Mu
;
資訊工程學系
;
工業工程與管理學系
The imbalanced data problem occurs in many application domains and is considered to be a challenging problem in machine learning and data mining. Most resampling methods for synthetic data focus on minority class without considering the data distribution of major classes. In contrast to previous works, the proposed method considers both majority classes and minority classes to learn feature embeddings and utilizes appropriate loss functions to make feature embedding as discriminative as possible. The proposed method is a comprehensive framework and different deep learning feature extractors can be utilized for different domains. We conduct experiments utilizing seven numerical datasets and one image dataset based on multiclass classification tasks. The experimental results indicate that the proposed method provides accurate and stable results.
DEM-Aided Block Adjustment for Satellite Images With Weak Convergence Geometry
(2010-04-01)
Teo, Tee-Ann
;
Chen, Liang-Chien
;
Liu, Chien-Liang
;
Tung, Yi-Chung
;
Wu, Wan-Yu
To acquire the largest possible coverage for environmental monitoring, it is important in most situations that the overlapping areas and the convergent angles of respective satellite images be small. The traditional bundle adjustment method used in aerial photogrammetry may not be the most suitable for direct orientation modeling in situations characterized by weak convergence geometry. We propose and compare three block adjustment methods for the processing of satellite images using the digital elevation model (DEM) as the elevation control. The first of these methods is a revised traditional bundle adjustment approach. The second is based on the direct georeferencing approach. The third is a rational function model with sensor-oriented rational polynomial coefficients. A collocation technique is integrated into all three methods to improve the positioning accuracy. Experimental results indicate that using the DEM as an elevation control can significantly improve the geometric accuracy as well as the geometric discrepancies between images. This is the case for all three methods. Moreover, the geometric performance of the three methods is similar. There is a significant improvement in geometric consistency between overlapping SPOT images with respect to single image adjustment for steep areas.
Enhancing Text Classification with the Universum
(2016)
Liu, Chien-Liang
;
Lee, Ching-Hsien
;
工業工程與管理學系
;
Department of Industrial Engineering and Management
The Universum is a data set that shares the same domain as the target problem, but does not comprise any category of interest. Recently, the concept of inference through contradictions has shown that the Universum provides a means for machine learning algorithms to encode prior knowledge into the model to improve performance. This work investigates whether text classification algorithms can benefit from the Universum when one has only a few labeled examples at hand. Additionally, this work proposes a confidence scheme to incorporate Universum into the learning process, and further devises a learning with Universum algorithm called Universum logistic regression (U-LR). The confidence scheme provides another means for machine learning algorithms to incorporate Universum into learning process. We conduct experiments on three data sets with several combinations. The experimental results indicate that the proposed method outperforms the other learning with Universum methods.
Epileptic Seizure Prediction With Multi-View Convolutional Neural Networks
(2019-01-01)
Liu, Chien-Liang
;
Xiao, Bin
;
Hsaio, Wen-Hoar
;
Tseng, Vincent S.
;
交大名義發表
The unpredictability of seizures is often considered by patients to be the most problematic aspect of epilepsy, so this work aims to develop an accurate epilepsy seizure predictor, making it possible to enable devices to warn patients of impeding seizures. To develop a model for seizure prediction, most studies relied on Electroencephalograms (EEGs) to capture physiological measurements of epilepsy. This work uses the two domains of EEGs, including frequency domain and time domain, to provide two different views for the same data source. Subsequently, this work proposes a multi-view convolutional neural network framework to predict the occurrence of epilepsy seizures with the goal of acquiring a shared representation of time-domain and frequency-domain features. By conducting experiments on Kaggle data set, we demonstrated that the proposed method outperforms all methods listed in the Kaggle leader board. Additionally, our proposed model achieves average area under the curve (AUCs) of 0.82 and 0.89 on two subjects of CHB-MIT scalp EEG data set. This work serves as an effective paradigm for applying deep learning approaches to the crucial topic of risk prediction in health domains.
Epileptic Seizure Prediction With Multi-View Convolutional Neural Networks
(2019-01-01)
Liu, Chien-Liang
;
Xiao, Bin
;
Hsaio, Wen-Hoar
;
Tseng, Vincent S.
;
交大名義發表
The unpredictability of seizures is often considered by patients to be the most problematic aspect of epilepsy, so this work aims to develop an accurate epilepsy seizure predictor, making it possible to enable devices to warn patients of impeding seizures. To develop a model for seizure prediction, most studies relied on Electroencephalograms (EEGs) to capture physiological measurements of epilepsy. This work uses the two domains of EEGs, including frequency domain and time domain, to provide two different views for the same data source. Subsequently, this work proposes a multi-view convolutional neural network framework to predict the occurrence of epilepsy seizures with the goal of acquiring a shared representation of time-domain and frequency-domain features. By conducting experiments on Kaggle data set, we demonstrated that the proposed method outperforms all methods listed in the Kaggle leader board. Additionally, our proposed model achieves average area under the curve (AUCs) of 0.82 and 0.89 on two subjects of CHB-MIT scalp EEG data set. This work serves as an effective paradigm for applying deep learning approaches to the crucial topic of risk prediction in health domains.
Fast recommendation on latent collaborative relations
(2016-10-01)
Liu, Chien-Liang
;
Wu, Xuan-Wei
;
資訊工程學系
;
工業工程與管理學系
;
Department of Computer Science
One important property of collaborative filtering recommender systems is that popular items are recommended disproportionately often because they provide extensive usage data and, thus, can be recommended to more users. Compared to popular products, the niches can be as economically attractive as mainstream fare for online retailers. The online retailers can stock virtually everything, and the number of available niche products exceeds the hits by several orders of magnitude. This work addresses accuracy, coverage and prediction time issues to propose a novel latent factor model called latent collaborative relations (LCR), which transforms the recommendation problem into a nearest neighbor search problem by using the proposed scoring function. We project users and items to the latent space, and calculate their similarities based on Euclidean metric. Additionally, the proposed model provides an elegant way to incorporate with locality sensitive hashing (LSH) to provide a fast recommendation while retaining recommendation accuracy and coverage. The experimental results indicate that the speedup is significant, especially when one is confronted with large-scale data sets. As for recommendation accuracy and coverage, the proposed method is competitive on three data sets. (C) 2016 Elsevier B.V. All rights reserved.
Hierarchical Hypothesis Structure for Ensemble Learning
(2017-01-01)
Yu, Chu-En
;
Liu, Chien-Liang
;
Hsieh, Hsin-Lung
;
資訊工程學系
;
工業工程與管理學系
One of the goals for the machine learning research is to improve the accuracy of the classification. Many research studies have focused on developing novel algorithms according to problem domains and statistical learning theory to continuously improve classification performance over the past decades. Recently, many researchers have found that performance bottleneck often occurs when only using a single classification algorithm, since each algorithm has its strength, but it also has its weakness. Ensemble learning, which combines several classifiers or hypotheses to become a strong classifier or learner, relies on the combination of various hypotheses rather than using state-of-the-art algorithms. In ensemble learning, hypothesis selection is crucial to performance, and the diversity of the selected hypotheses is an important selection criterion. This work proposes three algorithms focusing on generating a hierarchical hypothesis structure to achieve the goal of hypothesis selection, in which the two hypotheses are combined based on particular criterion. We conduct experiments on 8 data sets, and the experimental results indicate that the proposed method outperforms random forest, which is a state-of-the-art method.
High-resolution depth and coastline over major atolls of South China Sea from satellite altimetry and imagery
(2016-04)
Hsiao, Yu-Shen
;
Hwang, Cheinway
;
Cheng, Yung-Sheng
;
Chen, Liang-Chien
;
Hsu, Hung-Jui
Current bathymetric models for the South China Sea (SCS) are largely based on predicted depths from gravity and sparse single-beam echo-sounder measurements. Such models lack high-resolution coastlines and shallow water bottom features around atolls and islands. This study refines the gravity field of the SCS using sea surface heights from measurements of satellite altimeter Geosat/GM, ERS-1/GM, Jason-1/GM and the original Cryosat-2. A new one-minute gravity anomaly grid is determined. The modeled gravity anomalies show a 6-mgal RMS discrepancy with shipborne measurements in shallow waters. An altimeter-only bathymetric model is derived from the new gravity grid by the gravity-geological method that uses the latest global and regional models of the ocean depth and marine gravity as a priori knowledge. The new model outperforms current SCS bathymetric models and is accurate to 100 m, based on comparison with multi-beam depth measurements. Optical images from IKONOS-2, QuickBird-2, GeoEye-1, WorldView-1-2 and -3, are rectified and digitized to derive the zero (coastline) and 20-m depth contours (reef lines) around 44 atolls, which are integrated with the altimeter-only depths, giving significantly improved accuracies and spatial resolutions in modeled depths. The improvement percentages of coastlines by the satellite imagery range from 50% to 97% at 41 of the 44 atolls. We establish a webpage for free access to the optical and depth images, and the depth and gravity grids. We will continue to update satellite images, altimeter-derived gravity grids and bathymetric models over major atolls of the SCS. (C) 2016 Elsevier Inc. All rights reserved.