Pub Date : 2023-02-20DOI: 10.1109/ICAIIC57133.2023.10067124
Sangwon Oh, Hyeju Shin, Minsoo Hahn, Jinsul Kim
With the emergence of a flexible mix of private and public clouds based on business requirements, the need for a system that supports application deployment to a variety of cloud environments has emerged. In particular, it is necessary to secure the security of data in applications based on federated learning and to monitor resource usage in the cloud. This paper seeks ways to monitor and manage cloud resource usage according to various hyperparameters when conducting federated learning in a hybrid cloud environment. In a Docker-based cloud environment, we present an improved method for using efficient cloud resources while controlling the metric and resource usage trend of the federated learning model according to the imbalance of the data set.
{"title":"Analysis of Resource Usage Management Plan for Federated Learning in Hybrid Cloud","authors":"Sangwon Oh, Hyeju Shin, Minsoo Hahn, Jinsul Kim","doi":"10.1109/ICAIIC57133.2023.10067124","DOIUrl":"https://doi.org/10.1109/ICAIIC57133.2023.10067124","url":null,"abstract":"With the emergence of a flexible mix of private and public clouds based on business requirements, the need for a system that supports application deployment to a variety of cloud environments has emerged. In particular, it is necessary to secure the security of data in applications based on federated learning and to monitor resource usage in the cloud. This paper seeks ways to monitor and manage cloud resource usage according to various hyperparameters when conducting federated learning in a hybrid cloud environment. In a Docker-based cloud environment, we present an improved method for using efficient cloud resources while controlling the metric and resource usage trend of the federated learning model according to the imbalance of the data set.","PeriodicalId":105769,"journal":{"name":"2023 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"41 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125011630","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}
Pub Date : 2023-02-20DOI: 10.1109/ICAIIC57133.2023.10067075
H. Honda, Phuong Dinh, Pham Thu Thao, Yuho Tabata, Bui Duc Anh
A novel non-cooperative game theory-based approach for dimensionality reduction is proposed. We regard the sample elements in a higher-dimensional space as players in a game each of which has its strategy. A set of these strategies was implemented as an embedding of dimensionality reduction, which maps the sample elements into lower-dimensional spaces. Based on the theory of non-cooperative $N$-player games, we show the existence of Nash equilibria. We also provide an algorithm that yields Nash equilibrium based on the theory of nonlinear functional analysis.
{"title":"Dimensionality reduction as a non-cooperative game","authors":"H. Honda, Phuong Dinh, Pham Thu Thao, Yuho Tabata, Bui Duc Anh","doi":"10.1109/ICAIIC57133.2023.10067075","DOIUrl":"https://doi.org/10.1109/ICAIIC57133.2023.10067075","url":null,"abstract":"A novel non-cooperative game theory-based approach for dimensionality reduction is proposed. We regard the sample elements in a higher-dimensional space as players in a game each of which has its strategy. A set of these strategies was implemented as an embedding of dimensionality reduction, which maps the sample elements into lower-dimensional spaces. Based on the theory of non-cooperative $N$-player games, we show the existence of Nash equilibria. We also provide an algorithm that yields Nash equilibrium based on the theory of nonlinear functional analysis.","PeriodicalId":105769,"journal":{"name":"2023 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"41 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123414365","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}
Pub Date : 2023-02-20DOI: 10.1109/ICAIIC57133.2023.10067101
M. Nur, Hazriani, N. K. Nur
This study aims to implement the genetic algorithm by testing the appropriate crossover methods in order to obtain optimal disaster evacuation routes based three main indicators, namely travel time, possible transportation mode, and affected road conditions. The research phase begins with establishing a flood-affected area scenario consisting of the victim's initial location, evacuation location, routing areas, affected road conditions, distance, as well as travel time. The genetic algorithm is applied by representing the genes and chromosomes based on the available data, generating the initial population and calculating the fitness value. At the stage of determining the parent in forming a new individual, roulette wheel selection is used. For the crossover method to produce new individuals, there are 3 methods tested namely single-point, two-point and uniform crossover. The new formed individuals are then mutated with a probability level of 0.1. The last stage is to form a new population by sorting individuals with the highest fitness value. These processes took place with an iteration limit of 1000. Based on the results of the implementation and tests conducted, the uniform crossover method has the most optimal results with accuracy 90% and highest fitness value of 0.896. Meanwhile, the two others methods two-point and single-point have extremely lower accuracy which are 70% and 60% respectively. This result confirmed the statement of previous research which convinced that the uniform crossover is the most effective crossover method.
{"title":"Crossover Methods Comparison in Flood Evacuation Route Optimization","authors":"M. Nur, Hazriani, N. K. Nur","doi":"10.1109/ICAIIC57133.2023.10067101","DOIUrl":"https://doi.org/10.1109/ICAIIC57133.2023.10067101","url":null,"abstract":"This study aims to implement the genetic algorithm by testing the appropriate crossover methods in order to obtain optimal disaster evacuation routes based three main indicators, namely travel time, possible transportation mode, and affected road conditions. The research phase begins with establishing a flood-affected area scenario consisting of the victim's initial location, evacuation location, routing areas, affected road conditions, distance, as well as travel time. The genetic algorithm is applied by representing the genes and chromosomes based on the available data, generating the initial population and calculating the fitness value. At the stage of determining the parent in forming a new individual, roulette wheel selection is used. For the crossover method to produce new individuals, there are 3 methods tested namely single-point, two-point and uniform crossover. The new formed individuals are then mutated with a probability level of 0.1. The last stage is to form a new population by sorting individuals with the highest fitness value. These processes took place with an iteration limit of 1000. Based on the results of the implementation and tests conducted, the uniform crossover method has the most optimal results with accuracy 90% and highest fitness value of 0.896. Meanwhile, the two others methods two-point and single-point have extremely lower accuracy which are 70% and 60% respectively. This result confirmed the statement of previous research which convinced that the uniform crossover is the most effective crossover method.","PeriodicalId":105769,"journal":{"name":"2023 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"41 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127659054","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}
Pub Date : 2023-02-20DOI: 10.1109/ICAIIC57133.2023.10067056
D. Lakew, Anh-Tien Tran, Arooj Masood, Nhu-Ngoc Dao, Sungrae Cho
Aerial access networks (AANs) consisting of low altitude platforms (LAPs) and high altitude platforms (HAPs) have been considered as emerging wireless networking technologies to enhance both the capacity and coverage of future wireless networks, especially in remote and hard to reach areas with lack of terrestrial base stations. However, the limited onboard resources and high dynamicity of the network make challenging to optimally manage both the communication and computation resources for an efficient aerial networking infrastructure. On the other hand, artificial intelligence (AI), especially reinforcement learning- and deep reinforcement learning-based networking, are attracting significant attention to capture the network dynamicity and long-term resource management performance, recently. Thus, in this paper, we first provide a taxonomy of AI-driven aerial access networks and then, present a review and discussion on the state-of-the-art researches on AI-driven AANs from the communication and computation perspective. Moreover, we identify existing research challenges and provide future research direction for further investigations.
{"title":"A Review on AI-Driven Aerial Access Networks: Challenges and Open Research Issues","authors":"D. Lakew, Anh-Tien Tran, Arooj Masood, Nhu-Ngoc Dao, Sungrae Cho","doi":"10.1109/ICAIIC57133.2023.10067056","DOIUrl":"https://doi.org/10.1109/ICAIIC57133.2023.10067056","url":null,"abstract":"Aerial access networks (AANs) consisting of low altitude platforms (LAPs) and high altitude platforms (HAPs) have been considered as emerging wireless networking technologies to enhance both the capacity and coverage of future wireless networks, especially in remote and hard to reach areas with lack of terrestrial base stations. However, the limited onboard resources and high dynamicity of the network make challenging to optimally manage both the communication and computation resources for an efficient aerial networking infrastructure. On the other hand, artificial intelligence (AI), especially reinforcement learning- and deep reinforcement learning-based networking, are attracting significant attention to capture the network dynamicity and long-term resource management performance, recently. Thus, in this paper, we first provide a taxonomy of AI-driven aerial access networks and then, present a review and discussion on the state-of-the-art researches on AI-driven AANs from the communication and computation perspective. Moreover, we identify existing research challenges and provide future research direction for further investigations.","PeriodicalId":105769,"journal":{"name":"2023 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127891959","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}
Pub Date : 2023-02-20DOI: 10.1109/ICAIIC57133.2023.10066985
D. M. S. Bhatti, Haewoon Nam
Federated learning is a novel approach of training the global model on the server by utilizing the personal data of the end users while data privacy is preserved. The users called clients are required to perform the local training using their local datasets and forward those trained local models to the server, in which the local models are aggregated to update the global model. This process of global training is carried out for several rounds until the convergence. Practically, the clients' data is non-independent and identically distributed (Non-IID). Hence, the updated local model of each client may vary from every other client due to heterogeneity among them. Hence, the process of aggregating the diversified local models of clients has a huge impact on the performance of global training. This article proposes a performance efficient aggregation approach for federated learning, which considers the data heterogeneity among clients before aggregating the received local models. The proposed approach is compared with the conventional federated learning methods, and it achieves improved performance.
{"title":"A Performance Efficient Approach of Global Training in Federated Learning","authors":"D. M. S. Bhatti, Haewoon Nam","doi":"10.1109/ICAIIC57133.2023.10066985","DOIUrl":"https://doi.org/10.1109/ICAIIC57133.2023.10066985","url":null,"abstract":"Federated learning is a novel approach of training the global model on the server by utilizing the personal data of the end users while data privacy is preserved. The users called clients are required to perform the local training using their local datasets and forward those trained local models to the server, in which the local models are aggregated to update the global model. This process of global training is carried out for several rounds until the convergence. Practically, the clients' data is non-independent and identically distributed (Non-IID). Hence, the updated local model of each client may vary from every other client due to heterogeneity among them. Hence, the process of aggregating the diversified local models of clients has a huge impact on the performance of global training. This article proposes a performance efficient aggregation approach for federated learning, which considers the data heterogeneity among clients before aggregating the received local models. The proposed approach is compared with the conventional federated learning methods, and it achieves improved performance.","PeriodicalId":105769,"journal":{"name":"2023 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"70 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121142592","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}
Pub Date : 2023-02-20DOI: 10.1109/ICAIIC57133.2023.10067006
J. Tsiligaridis
An ensemble method is viewed as a compound model. The purpose of such a model is to achieve better predictive performance. The attempt is to tune predictions to observations by decreasing model variance, and bias. First the work focuses at the presentation of the Projective Decision Tree Algorithm (PA), a sort of Decision Tree (DT) based on purity and using the criterion of next node (CNN). Secondly, two sets of algorithms that provide improvement of the predictive performance are developed the first set of the Tree-Based Ensemble models of bagging and boosting types and the second set of known individual algorithms. The accuracy performance of the two sets with comparison is examined. Promising results based on accuracy of the proposed models are obtained.
{"title":"Tree-Based Ensemble Models and Algorithms for Classification","authors":"J. Tsiligaridis","doi":"10.1109/ICAIIC57133.2023.10067006","DOIUrl":"https://doi.org/10.1109/ICAIIC57133.2023.10067006","url":null,"abstract":"An ensemble method is viewed as a compound model. The purpose of such a model is to achieve better predictive performance. The attempt is to tune predictions to observations by decreasing model variance, and bias. First the work focuses at the presentation of the Projective Decision Tree Algorithm (PA), a sort of Decision Tree (DT) based on purity and using the criterion of next node (CNN). Secondly, two sets of algorithms that provide improvement of the predictive performance are developed the first set of the Tree-Based Ensemble models of bagging and boosting types and the second set of known individual algorithms. The accuracy performance of the two sets with comparison is examined. Promising results based on accuracy of the proposed models are obtained.","PeriodicalId":105769,"journal":{"name":"2023 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115964431","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}
Pub Date : 2023-02-20DOI: 10.1109/ICAIIC57133.2023.10066972
Sunghoon Hong, Dae-Geun Park
Convolutional neural networks with powerful visual image analysis for artificial intelligence are gaining popularity in many research fields, leading to the development of various high-performance algorithms for convolution operators present in these networks. One of these approaches is implemented with general matrix multiplication (GEMM) using the well-known im2col transform for fast convolution operations. In this paper, we propose a multi-core processor-based convolution technique for high-speed convolutional neural networks (CNNs) using differential images. The proposed method improves the convolutional layer's response speed by reducing the computational complexity and using multi-thread technology. In addition, the proposed algorithm has the advantage of being compatible with all types of CNNs. We use the darknet network to evaluate the convolutional layer's performance and show the best performance of the proposed algorithm when using 4-thread parallel processing.
{"title":"Differential Image-based Fast and Compatible Convolutional Layers for Multi-core Processors","authors":"Sunghoon Hong, Dae-Geun Park","doi":"10.1109/ICAIIC57133.2023.10066972","DOIUrl":"https://doi.org/10.1109/ICAIIC57133.2023.10066972","url":null,"abstract":"Convolutional neural networks with powerful visual image analysis for artificial intelligence are gaining popularity in many research fields, leading to the development of various high-performance algorithms for convolution operators present in these networks. One of these approaches is implemented with general matrix multiplication (GEMM) using the well-known im2col transform for fast convolution operations. In this paper, we propose a multi-core processor-based convolution technique for high-speed convolutional neural networks (CNNs) using differential images. The proposed method improves the convolutional layer's response speed by reducing the computational complexity and using multi-thread technology. In addition, the proposed algorithm has the advantage of being compatible with all types of CNNs. We use the darknet network to evaluate the convolutional layer's performance and show the best performance of the proposed algorithm when using 4-thread parallel processing.","PeriodicalId":105769,"journal":{"name":"2023 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"59 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126403266","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}
Pub Date : 2023-02-20DOI: 10.1109/ICAIIC57133.2023.10067092
R. Lackes, J. Sengewald
Predicting the total manufacturing costs of a new product early in its development is an obstacle for many businesses, especially when selecting between different product designs and their cost implications. Typically, material costs comprise a large part of total manufacturing costs, and therefore obtaining an early estimate of material costs can help businesses in predicting the total manufacturing costs more accurately. At the early stage of product development, with many imponderables and frequent design modifications, it would be impractical to obtain quotations from suppliers. We, therefore, developed a two-stage machine learning scheme estimating the material cost to guide alternative product design choices that yield a lower total manufacturing cost. Our innovative two-stage technique for cost estimation is meant to overcome this issue. In this paper, we demonstrate that neural networks, a prevalent technique in the literature, can be enhanced by adding the concept of modularity to the estimation of the pricing of technical components already during the design process of a new product.
{"title":"Early Product Cost Estimation by Intelligent Machine Learning Algorithms","authors":"R. Lackes, J. Sengewald","doi":"10.1109/ICAIIC57133.2023.10067092","DOIUrl":"https://doi.org/10.1109/ICAIIC57133.2023.10067092","url":null,"abstract":"Predicting the total manufacturing costs of a new product early in its development is an obstacle for many businesses, especially when selecting between different product designs and their cost implications. Typically, material costs comprise a large part of total manufacturing costs, and therefore obtaining an early estimate of material costs can help businesses in predicting the total manufacturing costs more accurately. At the early stage of product development, with many imponderables and frequent design modifications, it would be impractical to obtain quotations from suppliers. We, therefore, developed a two-stage machine learning scheme estimating the material cost to guide alternative product design choices that yield a lower total manufacturing cost. Our innovative two-stage technique for cost estimation is meant to overcome this issue. In this paper, we demonstrate that neural networks, a prevalent technique in the literature, can be enhanced by adding the concept of modularity to the estimation of the pricing of technical components already during the design process of a new product.","PeriodicalId":105769,"journal":{"name":"2023 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"64 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126404925","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}
Pub Date : 2023-02-20DOI: 10.1109/ICAIIC57133.2023.10067123
Min Woo Yoo, Chaehyun Lee, Dong Seog Han
This paper proposes a system in which a golf cart recognizes and tracks a user using a deep learning algorithm. Existing tracking golf carts use image processing algorithms or wearable sensors. However, image processing algorithms have low user recognition and tracking capabilities. In addition, the recognition and tracking system using a wearable sensor has a problem that requires an additional wearable sensor. We propose a non-attached smart golf cart using a deep learning algorithm to solve this problem. Deep learning object detection and classification algorithms are used to detect people and hands and recognize gestures in the detected hands. The golf cart performs user recognition, tracking, and human vehicle interface(HVI) by using the box of people and hands and gesture information. This paper verifies the algorithm on the golf cart.
{"title":"Deep Learning-based Human Vehicle Interface for Smart Golf Cart","authors":"Min Woo Yoo, Chaehyun Lee, Dong Seog Han","doi":"10.1109/ICAIIC57133.2023.10067123","DOIUrl":"https://doi.org/10.1109/ICAIIC57133.2023.10067123","url":null,"abstract":"This paper proposes a system in which a golf cart recognizes and tracks a user using a deep learning algorithm. Existing tracking golf carts use image processing algorithms or wearable sensors. However, image processing algorithms have low user recognition and tracking capabilities. In addition, the recognition and tracking system using a wearable sensor has a problem that requires an additional wearable sensor. We propose a non-attached smart golf cart using a deep learning algorithm to solve this problem. Deep learning object detection and classification algorithms are used to detect people and hands and recognize gestures in the detected hands. The golf cart performs user recognition, tracking, and human vehicle interface(HVI) by using the box of people and hands and gesture information. This paper verifies the algorithm on the golf cart.","PeriodicalId":105769,"journal":{"name":"2023 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134243467","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}
Pub Date : 2023-02-20DOI: 10.1109/ICAIIC57133.2023.10067074
A. Lit, Jamirin Shaet Joshima, S. Suhaili, N. Rajaee, S. K. Sahari, R. Sapawi
Due to its unique capability to communicate with long-distance communication processing cores in a single hop, on-chip wireless channels are utilized to reduce the network latency between the distant processing cores. Thus, due to its CMOS compatibility and architectural adaptability, wireless network-on-chip (NoC) is envisaged as a complement to the traditional NoC, which is attractive as wireless transmission will not require a wiring infrastructure. This paper evaluates three different deterministic routing algorithms (XY, west-first, and north-last) on a 100-core mesh WiNoC architecture. There are four wireless hubs equally located for each subnet on the mesh WiNoC architecture to examine its global transmission latency, throughput, and energy characteristics. In addition, the cycle-accurate Noxim simulator is employed to carry out the simulation for the WiNoC infrastructure under test using random and transpose traffic workload distribution. Experimental results show that, under a random traffic scenario, the XY routing algorithm provides the best packet injection rate (PIR) performance at 0.013 flits/cycle/tile. However, the investigated deterministic routing algorithms show no significant performance differences under the transpose traffic, as all of the routing approaches saturated at the same PIR point of 0.007 flit/cycle/tile.
{"title":"Evaluation of Deterministic Routing on 100-cores Mesh Wireless NoC","authors":"A. Lit, Jamirin Shaet Joshima, S. Suhaili, N. Rajaee, S. K. Sahari, R. Sapawi","doi":"10.1109/ICAIIC57133.2023.10067074","DOIUrl":"https://doi.org/10.1109/ICAIIC57133.2023.10067074","url":null,"abstract":"Due to its unique capability to communicate with long-distance communication processing cores in a single hop, on-chip wireless channels are utilized to reduce the network latency between the distant processing cores. Thus, due to its CMOS compatibility and architectural adaptability, wireless network-on-chip (NoC) is envisaged as a complement to the traditional NoC, which is attractive as wireless transmission will not require a wiring infrastructure. This paper evaluates three different deterministic routing algorithms (XY, west-first, and north-last) on a 100-core mesh WiNoC architecture. There are four wireless hubs equally located for each subnet on the mesh WiNoC architecture to examine its global transmission latency, throughput, and energy characteristics. In addition, the cycle-accurate Noxim simulator is employed to carry out the simulation for the WiNoC infrastructure under test using random and transpose traffic workload distribution. Experimental results show that, under a random traffic scenario, the XY routing algorithm provides the best packet injection rate (PIR) performance at 0.013 flits/cycle/tile. However, the investigated deterministic routing algorithms show no significant performance differences under the transpose traffic, as all of the routing approaches saturated at the same PIR point of 0.007 flit/cycle/tile.","PeriodicalId":105769,"journal":{"name":"2023 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":"4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132784390","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：