學術出版;;Publications

Permanent URI for this communityhdl:11536/1

收集所有交通大學研究人員與學生個人或團體所發表之各項學術著作，涵蓋期刊論文、教師專書、學位論文等項。;;Collections of Publications of NCTU Researchers and Students Like Journal Papers and Thesis.

Browse

🏛️國立陽明交通大學機構典藏

Search Results

Now showing 1 - 9 of 9

An Interpretable Compression and Classification System: Theory and Applications
(2020-01-01)
Tseng, Tzu-Wei
;
Yang, Kai-Jiun
;
Kuo, C-C Jay
;
Tsai, Shang-Ho
;
電子工程學系及電子研究所
This study proposes a low-complexity interpretable classification system. The proposed system contains main modules including feature extraction, feature reduction, and classification. All of them are linear. Thanks to the linear property, the extracted and reduced features can be inversed to original data, like a linear transform such as Fourier transform, so that one can quantify and visualize the contribution of individual features towards the original data. Also, the reduced features and reversibility naturally endure the proposed system ability of data compression. This system can significantly compress data with a small percent deviation between the compressed and the original data. At the same time, when the compressed data is used for classification, it still achieves high testing accuracy. Furthermore, we observe that the extracted features of the proposed system can be approximated to uncorrelated Gaussian random variables. Hence, classical theory in estimation and detection can be applied for classification. This motivates us to propose using a MAP (maximum a posteriori) based classification method. As a result, the extracted features and the corresponding performance have statistical meaning and mathematically interpretable. Simulation results show that the proposed classification system not only enjoys significant reduced training and testing time but also high testing accuracy compared to the conventional schemes.
Multiclass machine learning classification of functional brain images for Parkinson's disease stage prediction
(2020-10-01)
Huang, Guan-Hua
;
Lin, Chih-Hsuan
;
Cai, Yu-Ren
;
Chen, Tai-Been
;
Hsu, Shih-Yen
We analyzed a data set containing functional brain images from 6 healthy controls and 196 individuals with Parkinson's disease (PD), who were divided into five stages according to illness severity. The goal was to predict patients' PD illness stages by using their functional brain images. We employed the following prediction approaches: multivariate statistical methods (linear discriminant analysis, support vector machine, decision tree, and multilayer perceptron [MLP]), ensemble learning models (random forest [RF] and adaptive boosting), and deep convolutional neural network (CNN). For statistical and ensemble models, various feature extraction approaches (principal component analysis [PCA], multilinear PCA, intensity summary statistics [IStat], and Laws' texture energy measure) were employed to extract features, the synthetic minority over-sampling technique was used to address imbalanced data, and the optimal combination of hyperparameters was found using a grid search. For CNN modeling, we applied an image augmentation technique to increase and balance data sizes over different disease stages. We adopted transfer learning to incorporate pretrained VGG16 weights and architecture into the model fitting, and we also tested a state-of-the-art machine learning model that could automatically generate an optimal neural architecture. We found that IStat consistently outperformed other feature extraction approaches. MLP and RF were the analytic approaches with the highest prediction accuracy rate for multivariate statistical and ensemble learning models, respectively. Overall, the deep CNN model with pretrained VGG16 weights and architecture outperformed other approaches; it captured critical features from imaging, effectively distinguished between normal controls and patients with PD, and achieved the highest classification accuracy.
Comparison Training for Computer Chinese Chess
(2020-06-01)
Chen, Jr-Chang
;
Tseng, Wen-Jie
;
Wu, I-Chen
;
Wei, Ting-Han
;
資訊工程學系
This paper describes the application of modified comparison training for automatic feature weight tuning. The final objective is to improve the evaluation functions used in Chinese chess programs. First, we apply n-tuple networks to extract features. N-tuple networks require very little expert knowledge through its large numbers of features, while simultaneously allowing easy access. Second, we propose a modified comparison training into which tapered eval is incorporated. Experiments show that with the same features and the same Chinese chess program, the automatically tuned feature weights achieved a win rate of 86.58% against the hand-tuned features. The above trained version was then improved by adding additional features, most importantly n-tuple features. This improved version achieved a win rate of 81.65% against the trained version without additional features.
Eyeblink recognition improves fatigue prediction from single-channel forehead EEG in a realistic sustained attention task
(2020-06-01)
Ko, Li-Wei
;
Komarov, Oleksii
;
Lai, Wei-Kai
;
Liang, Wei-Gang
;
Jung, Tzyy-Ping
Objective. A passive brain-computer interface recognizes its operator's cognitive state without an explicitly performed control task. This technique is commonly used in conjunction with consumer-grade EEG devices for detecting the conditions of fatigue, attention, emotional arousal, or motion sickness. While it is easy to mount the sensors in the forehead area, which is not covered with hair, the recorded signals become greatly contaminated with eyeblink and movement artifacts, which makes it a challenge to acquire the data of suitable for analysis quality, particularly in few channel systems where a lack of spatial information limits the applicability of sophisticated signal cleaning algorithms. In this article, we demonstrate that by combining the features associated with electrocortical activities and eyeblink recognition analysis, it becomes feasible to design an accurate system for the inattention state prediction using just a single EEG sensor. Approach. Fifteen healthy 22-28 years old participants took part in the experiment that implemented a realistic sustained attention task of nighttime highway driving in a virtual environment. The EEG data were collected using a portable wireless Mindo-4 device, which constitutes an adjustable elastic strip with foam-based sensors, a data-acquisition module, an amplification and digitizing unit, and a Bluetooth (R) module. Main results. The spectral analysis of the EEG samples that immediately preceded the lane departure events revealed alterations in the tonic power spectral density, which accompanied elongations in the drivers' reaction times. The RMSE of the predicted reaction times, which are based on a combination of the brain-related and eyeblink features, is 0.034 +/- 0.019 s, and the r(2) is 0.885 +/- 0.057 according to a within-session leave-one-trial-out cross-validation. Significance. The drowsiness prediction from a frontal single-channel setup can achieve a comparable performance with using an array of occipital EEG sensors. As a direct result of utilizing a dry sensor placed in the non-covered with hair head area, the proposed approach in this study is low-cost and user-friendly.
Global Cortical Network Distinguishes Motor Imagination of the Left and Right Foot
(2020-01-01)
Phang, Chun-Ren
;
Ko, Li-Wei
;
交大名義發表
;
生物資訊及系統生物研究所
;
National Chiao Tung University
Conventional passive lower limb rehabilitation is suboptimal since the brain is not actively involved in the training. An autonomous motor imagery brain-computer interface (MI-BCI) could potentially improve rehabilitation outcomes. However, motor cortex regions associated with the individual feet are anatomically close to each other. This presents a difficulty in distinguishing the left and right foot MI during rehabilitation therapy. To overcome this difficulty, we extracted functional connectivity to measure the global cortical network via electroencephalography (EEG) signals. Fourteen spatial connections (P3-Fp1, P3-F3, P3-F7, P3-C3, T5-F7, T5-C3, T5-T3, Fp2-T5, Fp2-P3, T6-Fp2, T6-T4, Cz-Fp1, Cz-F7 and Fp2-F7) found across twelve subjects significantly differed between the left and right foot MI, evidencing nonlocalized brain activity during MI. Foot MI were distinguished using machine learning algorithms in terms of the time- and frequency-domain connectivities extracted from Pearson & x2019;s correlation, multivariate autoregression (MVAR), bandpass correlation, and partial directed coherence (PDC) models. The results showed that connectivity extracted by pairwise Pearson & x2019;s correlation could be distinguished with 86.26 & x00B1; 9.95 & x0025;, while in the frequency-domain, the gamma band presented the best classification accuracy of 73.55 & x00B1; 17.11 & x0025;. We attempted to simulate asynchronous real-time classification paradigms in order to evaluate the classification performance of connectivity features compared to common spatial pattern (CSP) and band power (BP). The results indicate correlation-connectivity has the best outcome, attaining an accuracy of 80.75 & x00B1; 9.51 & x0025; in asynchronous classification.
Big Data Platform Configuration Using Machine Learning
(2020-05-01)
Yeh, Chao-Chun
;
Lu, Han-Lin
;
Zhou, Jiazheng
;
Chang, Sheng-An
;
Lin, Xuan-Yi
By ensuring well-developed complex big data platform architectures, data engineers provide data scientists and analysts infrastructure with computational and storage resources to perform their research. Based on such supports, data scientists are provided an opportunity to focus on their domain problems and design the required intelligent modules (i.e., prepare the data; select, train, and tune the machine-learning modules; and validate the results). However, there are still gaps between system engineering and data scientist/engineering teams. Generally, system engineers have limited knowledge on the application domains and the purposes of an analytical program. On the contrary, both data scientists and engineers are usually unfamiliar with the configuration of a computational system, file system, and database. However, the performance of an application can be affected by a system's configuration, and the data scientists and engineers have little information and knowledge about which of the system's properties can affect the application's performance. As a typical example, for Internet-scale applications that have thousands of computing nodes or billions of Internet of Things devices, even a slight improvement may have an enormous influence on energy management and environmental protection issues. To bridge the gap between system engineering and data scientist/engineering teams, we proposed the concept of a configuration layer based on a big data platform, Hadoop. We built a configuration tuner, BigExplorer, to collect and preprocess data. Furthermore, we also created golden configurations for performance improvement. Based on the processed data, we used a semi-automatic feature engineering technique to provide more features for data engineers and developed the performance model using three different machine learning algorithms (i.e., random forest, gradient boosting machine, and support vector machine). Using the commonly used benchmarks of Word-Count, TeraSort, and Pig workloads, our configuration tuner achieved a significant performance improvement of 28%-51% for different workloads than using the rule-of-thumb configuration.
A Robust Algorithm for Predicting Attacks Using Collaborative Security Logs
(2020-05-01)
Rezapour, Amir
;
Tzeng, Wen-Guey
;
資訊工程學系
;
Department of Computer Science
As networks become ubiquitous in our daily lives, users rely more on networks for exchanging data and communication. However, numerous new and sophisticated attacks that endanger security of users have been reported. In practice, blacklisting illicit sources has been a fundamental defense strategy in recent years. In this paper, we propose a predictor that is based on the observations from a centralized log-sharing infrastructure. Our observations include the direct relation between attackers and victims, victim similarities, and attacker correlations. We compile a customized blacklist for each Dshield.org contributor using a weighted function of direct and indirect relations between victims and attackers. This list not only offers a significantly higher prediction ratio, but also includes source addresses with potentially higher threats. We evaluate our predictor using two months of malicious activities acquired from Dshield.org. The experimental results demonstrate a significant improvement over previous algorithms.
Optimizing Fixation Prediction Using Recurrent Neural Networks for 360 degrees Video Streaming in Head-Mounted Virtual Reality
(2020-03-01)
Fan, Ching-Ling
;
Yen, Shou-Cheng
;
Huang, Chun-Ying
;
Hsu, Cheng-Hsin
;
資訊工程學系
We study the problem of predicting the viewing probability of different parts of 360. videos when streaming them to head-mounted displays. We propose a fixation prediction network based on recurrent neural network, which leverages sensor and content features. The content features are derived by computer vision (CV) algorithms, which may suffer from inferior performance due to various types of distortion caused by diverse 360. video projection models. We propose a unified approach with overlapping virtual viewports to eliminate such negative effects, andwe evaluate our proposed solution using severalCValgorithms, such as saliency detection, face detection, and object detection. We find that overlapping virtual viewports increase the performance of these existing CV algorithms that were not trained for 360. videos. We next fine-tune our fixation prediction network with diverse design options, including: 1) with or without overlapping virtual viewports, 2) with or without future content features, and 3) different feature sampling rates. We empirically choose the best fixation prediction network and use it in a 360. video streaming system. We conduct extensive trace-driven simulations with a large-scale dataset to quantify the performance of the 360. video streaming system with different fixation prediction algorithms. The results show that our proposed fixation prediction network outperforms other algorithms in several aspects, such as: 1) achieving comparable video quality (average gaps between -0.05 and 0.92 dB), 2) consuming much less bandwidth (average bandwidth reduction by up to 8Mb/s), 3) reducing the rebuffering time (on average 40 s in bandwidth-limited 4G cellular networks), and 4) running in real-time (at most 124 ms).
Implementing AI as Cyber IoT Devices: The House Valuation Example
(2020-04-01)
Lin, Yun-Wei
;
Lin, Yi-Bing
;
Liu, Chun-You
;
Lin, Jiun-Yi
;
Shih, Yu-Lin
Internet of Things (IoT) has been widely utilized with artificial intelligence (AI). However, it requires substantial effort to integrate AI and big data with IoT. To mitigate this problem, AItalk was proposed. By treating AI as a cyber IoT device, we do not need to write code of the AI mechanism in the network applications, as traditional AI-based IoT applications did. This article describes how the AI tools such as scikit-learn and TensorLayer are accommodated as cyber IoT devices in AItalk, and extends AItalk for non-IoT applications. We use house valuation as an example to show how AItalk can flexibly include the factors that have significant impact on the house price. We show that by adding extra features other than housing profile features, the accuracy for the prediction (valuation) can be improved by 38 & x0025;. We also investigated the communication overhead of the distributed AItalk structure, which is 3.7 & x0025; for the computation of one house price valuation.

學術出版;;Publications

Browse

Filters

Settings

Sort By

Results per page

🏛️國立陽明交通大學機構典藏

Search Results