This study applies big data analysis techniques to analyze soccer managers' tactics and formations. For each playing position, the Boruta algorithm (a feature engineering algorithm) is applied to select the important features. K-means clustering was performed using the selected features, enabling the definition of the detailed roles of each position, such as holding midfielder and deep-lying playmaker. The analysis was conducted by dividing the CL (Champions League Level), EL (Europa League Level), ML (Middle Level) and RL (Relegation Level) to identify the differences in the tactics and formation patterns of the managers according to the level of opponent. Moreover, to include synergy between the players, weighted association rule mining was performed using the rating data as the weight to detect the strategy for each club. This implies that a manager establishes formations and tactics according to the level of the opponent.
{"title":"Discovering Synergic Association by Feature Clustering from Soccer Players","authors":"G. Lee, Gen Li, David Camacho, Jason J. Jung","doi":"10.1145/3400286.3418255","DOIUrl":"https://doi.org/10.1145/3400286.3418255","url":null,"abstract":"This study applies big data analysis techniques to analyze soccer managers' tactics and formations. For each playing position, the Boruta algorithm (a feature engineering algorithm) is applied to select the important features. K-means clustering was performed using the selected features, enabling the definition of the detailed roles of each position, such as holding midfielder and deep-lying playmaker. The analysis was conducted by dividing the CL (Champions League Level), EL (Europa League Level), ML (Middle Level) and RL (Relegation Level) to identify the differences in the tactics and formation patterns of the managers according to the level of opponent. Moreover, to include synergy between the players, weighted association rule mining was performed using the rating data as the weight to detect the strategy for each club. This implies that a manager establishes formations and tactics according to the level of the opponent.","PeriodicalId":326100,"journal":{"name":"Proceedings of the International Conference on Research in Adaptive and Convergent Systems","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-10-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130708527","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}
Recently, many companies have been introducing micro-service architecture framework technology to increase development productivity and portability. Google's Kubernetes is a representative micro-service architecture framework that introduces the concept of clusters to provide flexible expansion and services[1, 2]. Kubernetes applies technologies such as Scheduling and Load Balancing on a cluster-by-cluster basis, and has the flexibility to add nodes or move services within a single cluster. However, it does not provide multiple-cluster services scheduling, load balancing or flexible dynamic cluster add/delete technologies. Thus, OpenMCP(Open Multi-Cluster Container Platform) was designed for flexible expansion and service delivery between Kubernetes based Multi-Cluster. Key features of OpenMCP include multi-cluster resource collection, resource analysis, scheduling, load balancing, Auto Scaling, Cluster Synchronization, Dynamic Policy, Domain Name Server (DNS) management.
{"title":"Design and Implementation an OpenMCP distributed collaborative container platform for flexible scaling and service delivery","authors":"Chan-Hong Kim, J. An, Younghwan Kim","doi":"10.1145/3400286.3418278","DOIUrl":"https://doi.org/10.1145/3400286.3418278","url":null,"abstract":"Recently, many companies have been introducing micro-service architecture framework technology to increase development productivity and portability. Google's Kubernetes is a representative micro-service architecture framework that introduces the concept of clusters to provide flexible expansion and services[1, 2]. Kubernetes applies technologies such as Scheduling and Load Balancing on a cluster-by-cluster basis, and has the flexibility to add nodes or move services within a single cluster. However, it does not provide multiple-cluster services scheduling, load balancing or flexible dynamic cluster add/delete technologies. Thus, OpenMCP(Open Multi-Cluster Container Platform) was designed for flexible expansion and service delivery between Kubernetes based Multi-Cluster. Key features of OpenMCP include multi-cluster resource collection, resource analysis, scheduling, load balancing, Auto Scaling, Cluster Synchronization, Dynamic Policy, Domain Name Server (DNS) management.","PeriodicalId":326100,"journal":{"name":"Proceedings of the International Conference on Research in Adaptive and Convergent Systems","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-10-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132729601","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}
With the recent rapid development of computing power, interest in machine learning research on large data sets is increasing significantly. The machine learning is used in a wide variety of fields, from information retrieval, data mining, and speech recognition to human-computer interaction and application development by non-experts using machine learning platforms. However, there is not enough research on load balancing for distributed systems composed of heterogeneous servers with different performances and architectures that process machine learning tasks. Therefore, in this paper, we propose level hashing-based load balancing applicable to heterogeneous machine learning platforms. The proposed load balancing technique improves the execution time of all machine learning tasks in a machine learning platform by considering the characteristics of machine learning tasks and computing resources of each server.
{"title":"Load Balancing for Machine Learning Platform in Heterogeneous Distribute Computing Environment","authors":"Younggwan Kim, Jusuk Lee, Ajung Kim, Jiman Hong","doi":"10.1145/3400286.3418265","DOIUrl":"https://doi.org/10.1145/3400286.3418265","url":null,"abstract":"With the recent rapid development of computing power, interest in machine learning research on large data sets is increasing significantly. The machine learning is used in a wide variety of fields, from information retrieval, data mining, and speech recognition to human-computer interaction and application development by non-experts using machine learning platforms. However, there is not enough research on load balancing for distributed systems composed of heterogeneous servers with different performances and architectures that process machine learning tasks. Therefore, in this paper, we propose level hashing-based load balancing applicable to heterogeneous machine learning platforms. The proposed load balancing technique improves the execution time of all machine learning tasks in a machine learning platform by considering the characteristics of machine learning tasks and computing resources of each server.","PeriodicalId":326100,"journal":{"name":"Proceedings of the International Conference on Research in Adaptive and Convergent Systems","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-10-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115078335","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}
This paper discusses the multi-content disentanglement issue in unsupervised image transfer model. Image transfer based on generative model such as VAE1 or GAN2 can be defined as mapping data from source domain to target domain. Existing disentanglement methods have focused on separating elements of latent vector to distinguish content and style information from an image. However, since it has focused on extracting information from all pixels, it is hard to perform image transfer while controlling specific contents. To solve this problem, image transfer which is able to control a specific content disentanglement has been suggested recently. In this paper, by adapting the disentanglement concept to control various specific contents in a image, we propose a suitable architecture for image transfer task such as adding or subtracting multiple contents. In addition, we also propose an adversarially-learned auxiliary discriminator to further improve the quality of synthesized images from the multi-content disentanglement method. Based on the proposed method, we can generate images by controlling two contents from the CelebA dataset, and prove that we can attach specific content more clearly with auxiliary discriminator.
{"title":"Adversarially-learned Image Transfer Model for Multi-content Disentanglement","authors":"H. Seo, Jee-Hyong Lee","doi":"10.1145/3400286.3418250","DOIUrl":"https://doi.org/10.1145/3400286.3418250","url":null,"abstract":"This paper discusses the multi-content disentanglement issue in unsupervised image transfer model. Image transfer based on generative model such as VAE1 or GAN2 can be defined as mapping data from source domain to target domain. Existing disentanglement methods have focused on separating elements of latent vector to distinguish content and style information from an image. However, since it has focused on extracting information from all pixels, it is hard to perform image transfer while controlling specific contents. To solve this problem, image transfer which is able to control a specific content disentanglement has been suggested recently. In this paper, by adapting the disentanglement concept to control various specific contents in a image, we propose a suitable architecture for image transfer task such as adding or subtracting multiple contents. In addition, we also propose an adversarially-learned auxiliary discriminator to further improve the quality of synthesized images from the multi-content disentanglement method. Based on the proposed method, we can generate images by controlling two contents from the CelebA dataset, and prove that we can attach specific content more clearly with auxiliary discriminator.","PeriodicalId":326100,"journal":{"name":"Proceedings of the International Conference on Research in Adaptive and Convergent Systems","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-10-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"120859488","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}
Spiking neural network (SNNs) have been widely studied as an analysis model for human brain functioning. The energy-efficient nature of SNNs have attracted attentions of engineering researchers in deep neural networks. They sometimes need to have a tool that transforms SNNs to be executed in a deep learning framework. Due to inherent difference in their components for SNNs and deep neural networks, there are some inevitable restrictions in such transformations. This paper presents a new design and simulation environment for SNNs, which allows to build various architecture of SNNs and transforms them into computation graphs for execution. It supports several training algorithms for them. It exports their functionalities as APIs in Python with which the developers can build, train, and execute SNN models.
{"title":"Spiking Neural Network Transformer for Deploying into a Deep Learning Framework","authors":"C. Han, K. Lee","doi":"10.1145/3400286.3418272","DOIUrl":"https://doi.org/10.1145/3400286.3418272","url":null,"abstract":"Spiking neural network (SNNs) have been widely studied as an analysis model for human brain functioning. The energy-efficient nature of SNNs have attracted attentions of engineering researchers in deep neural networks. They sometimes need to have a tool that transforms SNNs to be executed in a deep learning framework. Due to inherent difference in their components for SNNs and deep neural networks, there are some inevitable restrictions in such transformations. This paper presents a new design and simulation environment for SNNs, which allows to build various architecture of SNNs and transforms them into computation graphs for execution. It supports several training algorithms for them. It exports their functionalities as APIs in Python with which the developers can build, train, and execute SNN models.","PeriodicalId":326100,"journal":{"name":"Proceedings of the International Conference on Research in Adaptive and Convergent Systems","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-10-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122707579","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}
In recent years, spiking neural networks (SNNs), a computing model inspired by the brain's ability to code and process information in the time domain with great computational power, has drawn a lot of attention from researchers for learning applications. For training in SNNs, several supervised spiking learning rules have been proposed, however, applying these learning algorithms to real-world problems yet remains an open issue. For this reason, this paper presents a new spiking neural network for the handwritten digit dataset classification problem. Our proposed network is trained by using the spike-based NormAD algorithm with a consistent winner-take-all mechanism. The experiment has shown a promising performance just after one epoch passing over the test dataset.
{"title":"Solving the Multi-class Classification Task in Spiking Neural Network by using Supervised Spiking Learning Rule with a Consistent Competitive Mechanism","authors":"Viet-Ngu Cong Huynh, K. Lee","doi":"10.1145/3400286.3418274","DOIUrl":"https://doi.org/10.1145/3400286.3418274","url":null,"abstract":"In recent years, spiking neural networks (SNNs), a computing model inspired by the brain's ability to code and process information in the time domain with great computational power, has drawn a lot of attention from researchers for learning applications. For training in SNNs, several supervised spiking learning rules have been proposed, however, applying these learning algorithms to real-world problems yet remains an open issue. For this reason, this paper presents a new spiking neural network for the handwritten digit dataset classification problem. Our proposed network is trained by using the spike-based NormAD algorithm with a consistent winner-take-all mechanism. The experiment has shown a promising performance just after one epoch passing over the test dataset.","PeriodicalId":326100,"journal":{"name":"Proceedings of the International Conference on Research in Adaptive and Convergent Systems","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-10-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121749942","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}
With the trend of 3D architecture and higher access rate, Phase Change Memory (PCM) storage devices face the overheating issue. This work is motivated by the observation that PCM devices might change states of memory cells with high temperature, and it will hurt the reliability of 3D PCM storage systems. Hence, we propose an Overheating-Avoidance Remapping Scheme (OARS) that controls the temperature of PCM layers and achieves wear-leveling of PCM cells inside PCM devices. Besides, we also take remapping overhead into consideration. The experiments were conducted based on the representative realistic workloads, and the results demonstrate the efficacy of the proposed scheme.
{"title":"Overheating-Avoidance Remapping Scheme for Reliability Enhancement of 3D PCM Storage Systems","authors":"Yu-Chen Lin, Tse-Yuan Wang, Che-Wei Tsao, Yuan-Hao Chang, Jian-Jia Chen, Xue Liu, Tei-Wei Kuo","doi":"10.1145/3400286.3418248","DOIUrl":"https://doi.org/10.1145/3400286.3418248","url":null,"abstract":"With the trend of 3D architecture and higher access rate, Phase Change Memory (PCM) storage devices face the overheating issue. This work is motivated by the observation that PCM devices might change states of memory cells with high temperature, and it will hurt the reliability of 3D PCM storage systems. Hence, we propose an Overheating-Avoidance Remapping Scheme (OARS) that controls the temperature of PCM layers and achieves wear-leveling of PCM cells inside PCM devices. Besides, we also take remapping overhead into consideration. The experiments were conducted based on the representative realistic workloads, and the results demonstrate the efficacy of the proposed scheme.","PeriodicalId":326100,"journal":{"name":"Proceedings of the International Conference on Research in Adaptive and Convergent Systems","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-10-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116952038","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}
M. Saad, Muhammad Toaha Raza Khan, M. Tariq, Dongkyun Kim
In the present era, internet of things (IoT) is prevailing very much in our daily life serving the concept of the smart applications, in which one can operate remote objects from a distant place. However, connectivity of the billions of devices has become a major concern in most of the prevailing researches. Massive connected devices used for smart applications consumes the network resources such as bandwidth and consumes the power to operate. Due to limited bandwidth, intermittent connectivity issues arises between smart devices which incorporates delay in the network. LoRaWAN (Long range Low power wide area network) developed by SemtechTM is a MAC layer protocol developed primarily for the IoT devices. In this paper, we implemented Long Short Term (LSTM) based smart gardening system, where end nodes collect the data from surrounding and sends to gateway using LoRa protocol. Edge Server is installed with the gateway on which LSTM based machine learning algorithm is running which predicts the future sensor values. For the predicted interval of time gateway sends the message to end nodes to remain inactive which saves the network bandwidth and also increases the life of sensors.
在当今时代，物联网(IoT)在我们的日常生活中非常盛行，服务于智能应用的概念，其中人们可以从遥远的地方操作远程对象。然而，数十亿设备的连接已成为大多数主流研究的主要关注点。智能应用中大量连接的设备消耗带宽等网络资源，也消耗运行的电力。由于带宽有限，智能设备之间出现间歇性连接问题，其中包含网络延迟。由SemtechTM开发的LoRaWAN(远程低功率广域网)是主要为物联网设备开发的MAC层协议。在本文中，我们实现了基于LSTM (Long Short Term)的智能园艺系统，在该系统中，终端节点通过LoRa协议从周围收集数据并发送到网关。边缘服务器安装了网关，在网关上运行基于LSTM的机器学习算法，该算法预测未来的传感器值。在预计的时间间隔内，网关向终端节点发送消息，使其处于非活动状态，从而节省了网络带宽，延长了传感器的寿命。
{"title":"LSTM Enabled Artificial Intelligent Smart Gardening System","authors":"M. Saad, Muhammad Toaha Raza Khan, M. Tariq, Dongkyun Kim","doi":"10.1145/3400286.3418260","DOIUrl":"https://doi.org/10.1145/3400286.3418260","url":null,"abstract":"In the present era, internet of things (IoT) is prevailing very much in our daily life serving the concept of the smart applications, in which one can operate remote objects from a distant place. However, connectivity of the billions of devices has become a major concern in most of the prevailing researches. Massive connected devices used for smart applications consumes the network resources such as bandwidth and consumes the power to operate. Due to limited bandwidth, intermittent connectivity issues arises between smart devices which incorporates delay in the network. LoRaWAN (Long range Low power wide area network) developed by SemtechTM is a MAC layer protocol developed primarily for the IoT devices. In this paper, we implemented Long Short Term (LSTM) based smart gardening system, where end nodes collect the data from surrounding and sends to gateway using LoRa protocol. Edge Server is installed with the gateway on which LSTM based machine learning algorithm is running which predicts the future sensor values. For the predicted interval of time gateway sends the message to end nodes to remain inactive which saves the network bandwidth and also increases the life of sensors.","PeriodicalId":326100,"journal":{"name":"Proceedings of the International Conference on Research in Adaptive and Convergent Systems","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-10-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124178244","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}
With the development of cloud technology, distributed and collaborative container platform technology has emerged to overcome the limitations of the existing stand-alone container platform, which has limitations in the mobility and resource scalability of cloud services. Distributed and collaborative container platform technology enables flexible expansion of resources and maximization of service mobility between container platforms distributed locally. In this paper, we propose a two-stage scheduler based on multi-resource metrics. The proposed scheduler determines the proper federated cluster where the request deployment can be deployed in a distributed and collaborative cluster environment. In order to select an proper federated cluster, filtering to select candidate clusters to which the scheduling request deployment can be deployed and scoring to evaluate the preference of each filtered cluster are performed.
{"title":"Scheduler for Distributed and Collaborative Container Clusters based on Multi-Resource Metric","authors":"Y. Lee, J. An, Younghwan Kim","doi":"10.1145/3400286.3418281","DOIUrl":"https://doi.org/10.1145/3400286.3418281","url":null,"abstract":"With the development of cloud technology, distributed and collaborative container platform technology has emerged to overcome the limitations of the existing stand-alone container platform, which has limitations in the mobility and resource scalability of cloud services. Distributed and collaborative container platform technology enables flexible expansion of resources and maximization of service mobility between container platforms distributed locally. In this paper, we propose a two-stage scheduler based on multi-resource metrics. The proposed scheduler determines the proper federated cluster where the request deployment can be deployed in a distributed and collaborative cluster environment. In order to select an proper federated cluster, filtering to select candidate clusters to which the scheduling request deployment can be deployed and scoring to evaluate the preference of each filtered cluster are performed.","PeriodicalId":326100,"journal":{"name":"Proceedings of the International Conference on Research in Adaptive and Convergent Systems","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-10-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129272952","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 PCIe (PCI Express) bus has long played a key role in interconnecting devices inside a system. In addition, advances in PCIe technology have made it possible to connect between servers using the PCIe bus. In this study, we've tried to improve the performance of our PCIe adapter cards for expanding the PCIe bus and connecting servers. Especially, we've looked for ways to make the most of the DMA capabilities offered by the PCIe switch chips mounted on our adapter cards. Our experimental results show that the dual ports method using multiple DMAs in each adapter card simultaneously, improves the performance up to 1.7 times compared to using a single port.
{"title":"Performance improvement of PCI Express adapter cards by adjusting the location of DMA functions","authors":"Kwangho Cha, Kyungmo Koo, Hyun Mi Jung","doi":"10.1145/3400286.3418229","DOIUrl":"https://doi.org/10.1145/3400286.3418229","url":null,"abstract":"The PCIe (PCI Express) bus has long played a key role in interconnecting devices inside a system. In addition, advances in PCIe technology have made it possible to connect between servers using the PCIe bus. In this study, we've tried to improve the performance of our PCIe adapter cards for expanding the PCIe bus and connecting servers. Especially, we've looked for ways to make the most of the DMA capabilities offered by the PCIe switch chips mounted on our adapter cards. Our experimental results show that the dual ports method using multiple DMAs in each adapter card simultaneously, improves the performance up to 1.7 times compared to using a single port.","PeriodicalId":326100,"journal":{"name":"Proceedings of the International Conference on Research in Adaptive and Convergent Systems","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2020-10-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129240839","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：