H. Nguyen, Nu-ri Shin, Gwanghyun Yu, Gyeong-Ju Kwon, Woon-Young Kwak, Jinyoung Kim
In this paper, the defective product classification based on deep learning for a smart factory is introduced. The proposed system contains PLC (Programmable Logic Controller), Artificial Intelligence (AI) embedded board and cloud service. The AI embedded board is connected and communicated to receive and send commands to PLC via SPI (Serial Peripheral Interface) protocol. The pre-trained defective product classification model is uploaded, saved on a cloud server and downloaded to AI Embedded board for each particular product. The core technique of the system is the AI-based embedded board. Due to the limitation of label data, we use transfer learning method to retrain deep neural networks (DNN). We implement and compare the classification results on different deep neural network including ResNet, DenseNet, and GoogLeNet. We trained these networks by GPU server on casting product classification data. After that, the pre-trained models are optimized and applied on practical embedded board. The experimental results show that our system is able to classify defective products with high accuracy and fast speed.
{"title":"Deep learning-based defective product classification system for smart factory","authors":"H. Nguyen, Nu-ri Shin, Gwanghyun Yu, Gyeong-Ju Kwon, Woon-Young Kwak, Jinyoung Kim","doi":"10.1145/3426020.3426039","DOIUrl":"https://doi.org/10.1145/3426020.3426039","url":null,"abstract":"In this paper, the defective product classification based on deep learning for a smart factory is introduced. The proposed system contains PLC (Programmable Logic Controller), Artificial Intelligence (AI) embedded board and cloud service. The AI embedded board is connected and communicated to receive and send commands to PLC via SPI (Serial Peripheral Interface) protocol. The pre-trained defective product classification model is uploaded, saved on a cloud server and downloaded to AI Embedded board for each particular product. The core technique of the system is the AI-based embedded board. Due to the limitation of label data, we use transfer learning method to retrain deep neural networks (DNN). We implement and compare the classification results on different deep neural network including ResNet, DenseNet, and GoogLeNet. We trained these networks by GPU server on casting product classification data. After that, the pre-trained models are optimized and applied on practical embedded board. The experimental results show that our system is able to classify defective products with high accuracy and fast speed.","PeriodicalId":305132,"journal":{"name":"The 9th International Conference on Smart Media and Applications","volume":"59 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127443360","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Geunha You, Gyoosik Kim, Jihyeon Park, Seong-je Cho, Minkyu Park
Code obfuscation is a technique that makes programs harder to understand. Malware writers widely the obfuscation technique to evade detection from anti-malware software, or to deter reverse engineering attempts for their malicious code. If we de-obfuscate the obfuscated code and restore it to the original code before obfuscation was applied, we can analyze the obfuscated malware effectively and efficiently. In this paper, we apply ReDex optimizer for reversing the control-flow obfuscation performed by the Obfuscapk system on open-source Android applications. We then analyze the effectiveness and limitations of ReDex in terms of its deobfuscation ability to reverse the control-flow obfuscation of Android apps. The experimental results show that ReDex can recover 1089 of 1108 apps obfuscated with control-flows obfuscation techniques of Obfuscapk obfuscator. During the process of optimizing bytecode, ReDex reduces the number of methods and fields significantly while it has a limitation in removing dead codes related to both useless goto statements and random nop instructions.
{"title":"Reversing Obfuscated Control Flow Structures in Android Apps using ReDex Optimizer","authors":"Geunha You, Gyoosik Kim, Jihyeon Park, Seong-je Cho, Minkyu Park","doi":"10.1145/3426020.3426089","DOIUrl":"https://doi.org/10.1145/3426020.3426089","url":null,"abstract":"Code obfuscation is a technique that makes programs harder to understand. Malware writers widely the obfuscation technique to evade detection from anti-malware software, or to deter reverse engineering attempts for their malicious code. If we de-obfuscate the obfuscated code and restore it to the original code before obfuscation was applied, we can analyze the obfuscated malware effectively and efficiently. In this paper, we apply ReDex optimizer for reversing the control-flow obfuscation performed by the Obfuscapk system on open-source Android applications. We then analyze the effectiveness and limitations of ReDex in terms of its deobfuscation ability to reverse the control-flow obfuscation of Android apps. The experimental results show that ReDex can recover 1089 of 1108 apps obfuscated with control-flows obfuscation techniques of Obfuscapk obfuscator. During the process of optimizing bytecode, ReDex reduces the number of methods and fields significantly while it has a limitation in removing dead codes related to both useless goto statements and random nop instructions.","PeriodicalId":305132,"journal":{"name":"The 9th International Conference on Smart Media and Applications","volume":"111 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126275912","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The power systems are deeply reforming to meet future power demands. With the continuous emergence of new technologies, the novel power system represented by microgrid has received more attention, and the research on the integration of emerging technologies of microgrid has become more focused. In this paper, a microgrid communication framework based on 5G technology is proposed, which makes full use of the low communication delay of 5G technology and the computation capacity of cloud/edge computing to implement the reconfiguration of microgrid deployed with DG(s). Lastly, we estimate the computing power of the cloud servers to predict the loads, and preprocess the restoration Optimal Configuration Table (OCT) scheme for instant fault restoration in the microgrid.
{"title":"Enhanced Microgrid Functions for Topology Reconfiguration and Fault Restoration","authors":"Xiancheng Wang, In-ho Ra","doi":"10.1145/3426020.3426165","DOIUrl":"https://doi.org/10.1145/3426020.3426165","url":null,"abstract":"The power systems are deeply reforming to meet future power demands. With the continuous emergence of new technologies, the novel power system represented by microgrid has received more attention, and the research on the integration of emerging technologies of microgrid has become more focused. In this paper, a microgrid communication framework based on 5G technology is proposed, which makes full use of the low communication delay of 5G technology and the computation capacity of cloud/edge computing to implement the reconfiguration of microgrid deployed with DG(s). Lastly, we estimate the computing power of the cloud servers to predict the loads, and preprocess the restoration Optimal Configuration Table (OCT) scheme for instant fault restoration in the microgrid.","PeriodicalId":305132,"journal":{"name":"The 9th International Conference on Smart Media and Applications","volume":"19 10 Suppl 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125993773","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Emotion information represents a user’s current emotional state and can be used in a variety of applications, such as cultural content services that recommend music according to user emotional states and user emotion monitoring. To increase user satisfaction, recommendation methods must understand and reflect user characteristics and circumstances, such as individual preferences and emotions. However, most recommendation methods do not reflect such characteristics accurately and are unable to increase user satisfaction. In this paper, six human emotions (neutral, happy, sad, angry, surprised, and bored) are broadly defined to consider user speech emotion information and recommend matching content. The “genetic algorithms as a feature selection method” (GAFS) algorithm was used to classify normalized speech according to speech emotion information. We used a support vector machine (SVM) algorithm and selected an optimal kernel function for recognizing the six target emotions. Performance evaluation results for each kernel function revealed that the radial basis function (RBF) kernel function yielded the highest emotion recognition accuracy of 86.98%. Additionally, content data (images and music) were classified based on emotion information using factor analysis, correspondence analysis, and Euclidean distance. Finally, speech information that was classified based on emotions and emotion information that was recognized through a collaborative filtering technique were used to predict user emotional preferences and recommend content that matched user emotions in a mobile application. To evaluate the performance of the proposed system, we performed verification based on the mean absolute error (MAE) metric. The evaluation results revealed an average accuracy of 87.43%.
情绪信息代表用户当前的情绪状态，可用于多种应用，例如根据用户情绪状态推荐音乐的文化内容服务和用户情绪监控。为了提高用户满意度，推荐方法必须理解和反映用户的特征和情况，如个人偏好和情绪。然而，大多数推荐方法不能准确反映这些特征，无法提高用户满意度。本文对人类的六种情绪(中性、快乐、悲伤、愤怒、惊讶和无聊)进行了广义的定义，以考虑用户的语音情绪信息并推荐匹配的内容。采用“遗传算法作为特征选择方法”(genetic algorithms as a feature selection method, GAFS)算法，根据语音情感信息对规范化语音进行分类。我们使用支持向量机(SVM)算法并选择一个最优核函数来识别六种目标情绪。各核函数的性能评价结果表明，径向基函数(RBF)核函数的情绪识别准确率最高，为86.98%。此外，基于情感信息，使用因子分析、对应分析和欧几里得距离对内容数据(图像和音乐)进行分类。最后，使用基于情绪分类的语音信息和通过协同过滤技术识别的情绪信息来预测用户的情绪偏好，并在移动应用程序中推荐与用户情绪匹配的内容。为了评估所提出系统的性能，我们基于平均绝对误差(MAE)度量进行了验证。评价结果显示平均准确率为87.43%。
{"title":"Emotion and Collaborative Filtering-Based Recommendation System","authors":"Tae-Yeun Kim, Sung-Hwan Kim","doi":"10.1145/3426020.3426119","DOIUrl":"https://doi.org/10.1145/3426020.3426119","url":null,"abstract":"Emotion information represents a user’s current emotional state and can be used in a variety of applications, such as cultural content services that recommend music according to user emotional states and user emotion monitoring. To increase user satisfaction, recommendation methods must understand and reflect user characteristics and circumstances, such as individual preferences and emotions. However, most recommendation methods do not reflect such characteristics accurately and are unable to increase user satisfaction. In this paper, six human emotions (neutral, happy, sad, angry, surprised, and bored) are broadly defined to consider user speech emotion information and recommend matching content. The “genetic algorithms as a feature selection method” (GAFS) algorithm was used to classify normalized speech according to speech emotion information. We used a support vector machine (SVM) algorithm and selected an optimal kernel function for recognizing the six target emotions. Performance evaluation results for each kernel function revealed that the radial basis function (RBF) kernel function yielded the highest emotion recognition accuracy of 86.98%. Additionally, content data (images and music) were classified based on emotion information using factor analysis, correspondence analysis, and Euclidean distance. Finally, speech information that was classified based on emotions and emotion information that was recognized through a collaborative filtering technique were used to predict user emotional preferences and recommend content that matched user emotions in a mobile application. To evaluate the performance of the proposed system, we performed verification based on the mean absolute error (MAE) metric. The evaluation results revealed an average accuracy of 87.43%.","PeriodicalId":305132,"journal":{"name":"The 9th International Conference on Smart Media and Applications","volume":"3 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129091523","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Duy Dao, Ngoc-Huynh Ho, Jahae Kim, Hyung-Jeong Yang
Alzheimer's Disease (AD) is known as a degenerative neurological progression that causes the loss of neurons and synapses in the cerebral cortex. The cognitive functions are gradually impaired over several to 20 years and no current cure. It is crucial for timely conversion prediction to AD in its earliest phrase. In this study, we propose a novel recurrent neural network (RNN) model to obtain biomarkers of brain and clinical diagnosis of each subject from only one to indefinitely forecast the biomarkers and clinical diagnosis at each timepoint in the future. However, unexpected missing observations is a common issue in longitudinal data. Moreover, in recurrent dynamical systems, gates should be closer to 1 to propagate more information from earlier time steps to later ones. Two strategies are explored to handle missing data and improve gating mechanisms in recurrent neural network to boost performance. Empirically, our gating mechanisms robustly improve the performance when longitudinal data is utilized. On the baseline dataset from The Alzheimer's Disease Prediction Of Longitudinal Evolution (TADPOLE) challenge, our proposal outperforms the conventional recurrent neural network in term of multi-class Area Under Curve (mAUC) which achieving 0.79734 ± 0.01785.
阿尔茨海默病(AD)是一种退化的神经系统疾病，会导致大脑皮层中神经元和突触的丧失。认知功能在几到20年内逐渐受损，目前尚无治愈方法。在广告投放的早期进行及时的转化预测是至关重要的。在这项研究中，我们提出了一种新的递归神经网络(RNN)模型，该模型可以从每个受试者的一个大脑生物标志物和临床诊断中获得未来每个时间点的生物标志物和临床诊断，从而无限期地预测未来每个时间点的生物标志物和临床诊断。然而，在纵向数据中，意想不到的缺失观测是一个常见的问题。此外，在循环动力系统中，门应该更接近于1，以便从较早的时间步向较晚的时间步传播更多的信息。为了提高递归神经网络的性能，本文探讨了两种策略来处理缺失数据和改进门控机制。经验表明，当纵向数据被利用时，我们的门控机制稳健地提高了性能。在the Alzheimer's Disease Prediction Of Longitudinal Evolution (TADPOLE) challenge的基线数据集上，我们的方法在多类曲线下面积(Area Under Curve, mAUC)方面优于传统的递归神经网络，达到0.79734±0.01785。
{"title":"Improving Recurrent Gate Mechanism For Time-to-Conversion Prediction Of Alzheimer's Disease","authors":"Duy Dao, Ngoc-Huynh Ho, Jahae Kim, Hyung-Jeong Yang","doi":"10.1145/3426020.3426036","DOIUrl":"https://doi.org/10.1145/3426020.3426036","url":null,"abstract":"Alzheimer's Disease (AD) is known as a degenerative neurological progression that causes the loss of neurons and synapses in the cerebral cortex. The cognitive functions are gradually impaired over several to 20 years and no current cure. It is crucial for timely conversion prediction to AD in its earliest phrase. In this study, we propose a novel recurrent neural network (RNN) model to obtain biomarkers of brain and clinical diagnosis of each subject from only one to indefinitely forecast the biomarkers and clinical diagnosis at each timepoint in the future. However, unexpected missing observations is a common issue in longitudinal data. Moreover, in recurrent dynamical systems, gates should be closer to 1 to propagate more information from earlier time steps to later ones. Two strategies are explored to handle missing data and improve gating mechanisms in recurrent neural network to boost performance. Empirically, our gating mechanisms robustly improve the performance when longitudinal data is utilized. On the baseline dataset from The Alzheimer's Disease Prediction Of Longitudinal Evolution (TADPOLE) challenge, our proposal outperforms the conventional recurrent neural network in term of multi-class Area Under Curve (mAUC) which achieving 0.79734 ± 0.01785.","PeriodicalId":305132,"journal":{"name":"The 9th International Conference on Smart Media and Applications","volume":"17 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129170214","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Cooperation or Cooperative behavior constrained between any two nodes or groups always result in constant scrutiny for reconfiguration. This continual reconfiguration creates a new modulus for expansion and thus detecting community structure can fundamentally become a problem of identifying groups and a leader in a network. In a network, the influencer is commonly termed as leader and the leader node is a node that has high attraction to increase, i.e., high degree of centrality. In this paper, we devised an efficient method to detect influencers in a network through cooperative and spread strategies. This dynamic strategy technique is used to detect subevents and anomalies through social and physical sensor data. This paper contributes toward a dynamic game theory approach for information maximization by maximizing the influence features over the network for higher information delivery over the dynamic network.
{"title":"Cooperative Influence Learning","authors":"Harshit Srivastava, In-ho Ra, R. Sankar","doi":"10.1145/3426020.3426159","DOIUrl":"https://doi.org/10.1145/3426020.3426159","url":null,"abstract":"Cooperation or Cooperative behavior constrained between any two nodes or groups always result in constant scrutiny for reconfiguration. This continual reconfiguration creates a new modulus for expansion and thus detecting community structure can fundamentally become a problem of identifying groups and a leader in a network. In a network, the influencer is commonly termed as leader and the leader node is a node that has high attraction to increase, i.e., high degree of centrality. In this paper, we devised an efficient method to detect influencers in a network through cooperative and spread strategies. This dynamic strategy technique is used to detect subevents and anomalies through social and physical sensor data. This paper contributes toward a dynamic game theory approach for information maximization by maximizing the influence features over the network for higher information delivery over the dynamic network.","PeriodicalId":305132,"journal":{"name":"The 9th International Conference on Smart Media and Applications","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129841173","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Parkinson’s disease (PD) is a common neurodegenerative disease that causes involuntary muscle movements and tremor among other symptoms. One approach to diagnosing this disease is by analyzing the electroencephalography (EEG) signals of the patients. However, due to the complexity of this type of signal, more advanced feature extraction and classification methods are required. The goal of this study is to combine six well-known features in EEG analysis with eight higher order statistical features and use them for classification of early stage PD (1st and 2nd stage) from a healthy control group. After extracting the required features, eight classifiers are employed to classify the signals.
{"title":"Early Stage Diagnosis of Parkinson’s Disease Using HOS Features of EEG Signals","authors":"S. A. Khoshnevis, In-ho Ra, R. Sankar","doi":"10.1145/3426020.3426160","DOIUrl":"https://doi.org/10.1145/3426020.3426160","url":null,"abstract":"Parkinson’s disease (PD) is a common neurodegenerative disease that causes involuntary muscle movements and tremor among other symptoms. One approach to diagnosing this disease is by analyzing the electroencephalography (EEG) signals of the patients. However, due to the complexity of this type of signal, more advanced feature extraction and classification methods are required. The goal of this study is to combine six well-known features in EEG analysis with eight higher order statistical features and use them for classification of early stage PD (1st and 2nd stage) from a healthy control group. After extracting the required features, eight classifiers are employed to classify the signals.","PeriodicalId":305132,"journal":{"name":"The 9th International Conference on Smart Media and Applications","volume":"22 22 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123423471","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Jihoon Lee, Seungmin Oh, Yeonggwang Kim, Dongsu Lee, Akm Ashiquzzaman, Jinsul Kim
Various smart farm technologies are currently being developed around the world to enhance agricultural competitiveness. Korea is also speeding up the development of Korean smart farm technology suitable for domestic environment, but it is difficult to develop high-reliability sensors and systems, and has problems such as preventing sensors from failing, so in this paper, environmental data values such as temperature, humidity, carbon dioxide, ammonia, etc. are sensed, refined, and pretreated to derive correlation and heat maps between sensors. This will not only predict the RUL (Remaining Useful Life) of the sensor using machine learning in the future, but also develop a reliable system by detecting failures and errors.
{"title":"Pig Farm Environment Sensor Data Correlation and Heatmap Analysis for Predicting Sensor Remaining Useful Life✱","authors":"Jihoon Lee, Seungmin Oh, Yeonggwang Kim, Dongsu Lee, Akm Ashiquzzaman, Jinsul Kim","doi":"10.1145/3426020.3426136","DOIUrl":"https://doi.org/10.1145/3426020.3426136","url":null,"abstract":"Various smart farm technologies are currently being developed around the world to enhance agricultural competitiveness. Korea is also speeding up the development of Korean smart farm technology suitable for domestic environment, but it is difficult to develop high-reliability sensors and systems, and has problems such as preventing sensors from failing, so in this paper, environmental data values such as temperature, humidity, carbon dioxide, ammonia, etc. are sensed, refined, and pretreated to derive correlation and heat maps between sensors. This will not only predict the RUL (Remaining Useful Life) of the sensor using machine learning in the future, but also develop a reliable system by detecting failures and errors.","PeriodicalId":305132,"journal":{"name":"The 9th International Conference on Smart Media and Applications","volume":"28 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114109500","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Many studies are being conducted recently on text-to-image and text-to-video generation based on deep-learning. Text-to-image generation shows a noticeable performance, while the performance related to text-to-video generation is yet insufficient. Because the results of text-to-video generation are derived as multiple images according to the sequence, more feature information needs to be considered than in the text-to-image generation, and the relevance of each image should be considered. Thus, this study proposes a method of a text-to-dynamic image generation focusing on character, temporal, and background information to consider the feature information of a video. This method can be used to quickly visualize ideas and produce prototypes during the production process of video materials such as advertisements, movies, and TV series, and to visualize and upload textual posts on video-based social media services such as TikTok, Instagram, and Flicker, or video-based platforms such as YouTube.
{"title":"A Text-to-Dynamic Image Generation Method using Feature Information of Video","authors":"Taekeun Hong, Kang-Hyo Kim, Kiho Lim, Pankoo Kim","doi":"10.1145/3426020.3426038","DOIUrl":"https://doi.org/10.1145/3426020.3426038","url":null,"abstract":"Many studies are being conducted recently on text-to-image and text-to-video generation based on deep-learning. Text-to-image generation shows a noticeable performance, while the performance related to text-to-video generation is yet insufficient. Because the results of text-to-video generation are derived as multiple images according to the sequence, more feature information needs to be considered than in the text-to-image generation, and the relevance of each image should be considered. Thus, this study proposes a method of a text-to-dynamic image generation focusing on character, temporal, and background information to consider the feature information of a video. This method can be used to quickly visualize ideas and produce prototypes during the production process of video materials such as advertisements, movies, and TV series, and to visualize and upload textual posts on video-based social media services such as TikTok, Instagram, and Flicker, or video-based platforms such as YouTube.","PeriodicalId":305132,"journal":{"name":"The 9th International Conference on Smart Media and Applications","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124266222","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Recent cloud IDE services provide containers as development environment to users. Since users have little knowledge on specific tasks to run and computing resources required in their containers, it is difficult to decide exactly how many containers to allocate to the cloud instance. Cloud services often employ conservative managing policy to make the cloud instances to an appropriate level, and only increase the instances little by little when their services encounter resource problems. In addition, a simple container placement policy creates a situation where no more containers can be allocated, even though resources are available in some of cloud instances depending on their execution situations. To improve this, we place as many cloud instances as possible based on the predicted container usage, which is collected from the usage data of the containers on previous cloud instances. When a cloud instance has too much surplus resource, we also employ container migration to effectively manage overall cloud instances. By equipping our cloud service with an intelligent management policy, we can reduce the total number of cloud instances in use and increase the cost efficiency for our cloud service by 14.7%, according to our simulation study.
{"title":"Cost-Effective Container Orchestration Using Usage Data","authors":"Y. Nam, Hwansoo Han","doi":"10.1145/3426020.3426123","DOIUrl":"https://doi.org/10.1145/3426020.3426123","url":null,"abstract":"Recent cloud IDE services provide containers as development environment to users. Since users have little knowledge on specific tasks to run and computing resources required in their containers, it is difficult to decide exactly how many containers to allocate to the cloud instance. Cloud services often employ conservative managing policy to make the cloud instances to an appropriate level, and only increase the instances little by little when their services encounter resource problems. In addition, a simple container placement policy creates a situation where no more containers can be allocated, even though resources are available in some of cloud instances depending on their execution situations. To improve this, we place as many cloud instances as possible based on the predicted container usage, which is collected from the usage data of the containers on previous cloud instances. When a cloud instance has too much surplus resource, we also employ container migration to effectively manage overall cloud instances. By equipping our cloud service with an intelligent management policy, we can reduce the total number of cloud instances in use and increase the cost efficiency for our cloud service by 14.7%, according to our simulation study.","PeriodicalId":305132,"journal":{"name":"The 9th International Conference on Smart Media and Applications","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129775893","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：