Pub Date : 2019-05-09DOI: 10.1109/ICNSC.2019.8743181
The Hung Pham, S. Mammar
This paper addresses the problem of attitude/altitude control of a quadrotor. The main contribution consists in developing a simple Linear Parameter Varying model which includes the motor dynamics and weight variations. Afterwards, a reference model is introduced and an error model is derived. An integral action is thus naturally included in the loop. The proposed controller takes the form of a static output feedback which is synthesised using the Linear Matrix Inequalities framework. Thanks to a relaxation method the nonlinear terms are removed from the matrix inequalities. The controller in then reconstructed as a combination of the integral of the error, the actual output and the preview reference signal. Simulations are conducted for a scenario showing the ability of the design method to handle different performance objectives.
{"title":"Quadrotor LPV Control using Static Output Feedback","authors":"The Hung Pham, S. Mammar","doi":"10.1109/ICNSC.2019.8743181","DOIUrl":"https://doi.org/10.1109/ICNSC.2019.8743181","url":null,"abstract":"This paper addresses the problem of attitude/altitude control of a quadrotor. The main contribution consists in developing a simple Linear Parameter Varying model which includes the motor dynamics and weight variations. Afterwards, a reference model is introduced and an error model is derived. An integral action is thus naturally included in the loop. The proposed controller takes the form of a static output feedback which is synthesised using the Linear Matrix Inequalities framework. Thanks to a relaxation method the nonlinear terms are removed from the matrix inequalities. The controller in then reconstructed as a combination of the integral of the error, the actual output and the preview reference signal. Simulations are conducted for a scenario showing the ability of the design method to handle different performance objectives.","PeriodicalId":291695,"journal":{"name":"2019 IEEE 16th International Conference on Networking, Sensing and Control (ICNSC)","volume":"39 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127125920","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 : 2019-05-09DOI: 10.1109/ICNSC.2019.8743270
Tong Zou, Yi Luo, Xiaohan Zhang, Qun Zhu, Yanlin He
For industry 4.0, intelligent modeling is very important. Modelling plays a very important role in making control strategies and production plans. Nevertheless, establishing an accurate and robust model becomes more difficult because of the increasing complexity of modelling data. To solve this problem, a novel feature extraction based multiple activation functions extreme learning machine (LV-MAFELM) is presented. The LV-MAFELM model is easy to construct: firstly, generate the input weights at random; secondly, select several different nonlinear activation functions and compute the hidden layer outputs; thirdly, extract principal components from the hidden layer outputs; finally, compute the output weights analytically. For verifying the model performance, the LV-MAFELM model is applied in one petrochemical industry process -the Purified Terephthalic Acid (PTA) process. Simulation results demonstrate that the presented LV-MAFELM achieves good performance, which indicates that accuracy and stability of energy prediction models can be improved.
{"title":"Intelligent modeling using a novel feature extraction based multiple activation functions extreme learning machine","authors":"Tong Zou, Yi Luo, Xiaohan Zhang, Qun Zhu, Yanlin He","doi":"10.1109/ICNSC.2019.8743270","DOIUrl":"https://doi.org/10.1109/ICNSC.2019.8743270","url":null,"abstract":"For industry 4.0, intelligent modeling is very important. Modelling plays a very important role in making control strategies and production plans. Nevertheless, establishing an accurate and robust model becomes more difficult because of the increasing complexity of modelling data. To solve this problem, a novel feature extraction based multiple activation functions extreme learning machine (LV-MAFELM) is presented. The LV-MAFELM model is easy to construct: firstly, generate the input weights at random; secondly, select several different nonlinear activation functions and compute the hidden layer outputs; thirdly, extract principal components from the hidden layer outputs; finally, compute the output weights analytically. For verifying the model performance, the LV-MAFELM model is applied in one petrochemical industry process -the Purified Terephthalic Acid (PTA) process. Simulation results demonstrate that the presented LV-MAFELM achieves good performance, which indicates that accuracy and stability of energy prediction models can be improved.","PeriodicalId":291695,"journal":{"name":"2019 IEEE 16th International Conference on Networking, Sensing and Control (ICNSC)","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122235061","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 : 2019-05-09DOI: 10.1109/ICNSC.2019.8743210
Qiaoli Zhuang, Dan You, Wenzhan Dai, Shouguang Wang, Jingiing Du
In deadtock prevention poticies based on siphon control the selection of a siphon to be controtted in each iteration may affect the structural comptexiiv and the behaviorat permissiveness of the controtted system In this paper, an iterative deadtock prevention policy based on mixed integer programming (MIP) is introducedfor a ctass of Petri nets called systems of sequential systems with shared resources (S4PR). Sonoe experinoents show that the resultant system obtained by the proposedpolicy has simpler structure and more permissive behavior than those obtained from existing noethods.
{"title":"An iterative Deadlock Prevention Policy Based on siphons","authors":"Qiaoli Zhuang, Dan You, Wenzhan Dai, Shouguang Wang, Jingiing Du","doi":"10.1109/ICNSC.2019.8743210","DOIUrl":"https://doi.org/10.1109/ICNSC.2019.8743210","url":null,"abstract":"In deadtock prevention poticies based on siphon control the selection of a siphon to be controtted in each iteration may affect the structural comptexiiv and the behaviorat permissiveness of the controtted system In this paper, an iterative deadtock prevention policy based on mixed integer programming (MIP) is introducedfor a ctass of Petri nets called systems of sequential systems with shared resources (S4PR). Sonoe experinoents show that the resultant system obtained by the proposedpolicy has simpler structure and more permissive behavior than those obtained from existing noethods.","PeriodicalId":291695,"journal":{"name":"2019 IEEE 16th International Conference on Networking, Sensing and Control (ICNSC)","volume":"201 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115889925","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 : 2019-05-09DOI: 10.1109/ICNSC.2019.8743326
Yi Du, Ting Zhang, Jiacun Wang
A new method for three-dimensional stochastic reconstruction of spatial data is proposed. This method introduces deep learning into the feature extraction and reconstruction process of discrete spatial data. In the training process, the spatial data features are learned by constructing a deep neural network, and the global correlation between data is obtained; then the reconstruction results are obtained by feature replication. In the training process, this method doesn’t need to scan the training image repeatedly, which is different from the traditional multiple-point simulation. The experimental results show that the structural features of reconstructed spatial data using this method are consistent with the training images.
{"title":"Discrete Spatial Data Reconstruction based on Deep Neural Network","authors":"Yi Du, Ting Zhang, Jiacun Wang","doi":"10.1109/ICNSC.2019.8743326","DOIUrl":"https://doi.org/10.1109/ICNSC.2019.8743326","url":null,"abstract":"A new method for three-dimensional stochastic reconstruction of spatial data is proposed. This method introduces deep learning into the feature extraction and reconstruction process of discrete spatial data. In the training process, the spatial data features are learned by constructing a deep neural network, and the global correlation between data is obtained; then the reconstruction results are obtained by feature replication. In the training process, this method doesn’t need to scan the training image repeatedly, which is different from the traditional multiple-point simulation. The experimental results show that the structural features of reconstructed spatial data using this method are consistent with the training images.","PeriodicalId":291695,"journal":{"name":"2019 IEEE 16th International Conference on Networking, Sensing and Control (ICNSC)","volume":"60 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114396599","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 : 2019-05-09DOI: 10.1109/ICNSC.2019.8743182
Kai Zhang, Q. Kang, Le Pan, X. Wang, Can Cui
One of the difficulties in computer vision is how to build an accurate classifier for a new target domain with insufficient labeled images from a related source domain with labeled images. Adversarial learning is a novel domain adaptation method that tackles this challenge by training robust deep networks and reducing the distribution difference between source and target domains, thus improving the classification performance on a target task. However, most prior adversarial adaptation learning approaches merely reduce the distribution difference across domains through GAN (Generative Adversarial Networks)-based loss, but when the performance of a generator or discriminator in GAN is degraded, the distribution difference between source and target domains are difficult to decrease. In this paper, we propose a novel generalized framework for adversarial domain adaptation, referred to as Generative Adversarial Distribution Matching. Our idea is to add the data discrepancy distance between source and target domains to the objective function of the generator so as to reduce distribution difference across domains through a Generator and a Discriminator compete against each other. Comprehensive experimental results confirm that it can well outperform several state-of-the-art methods for cross-domain image classification problems.
{"title":"A Generative Adversarial Distribution Matching Framework for Visual Domain Adaptation","authors":"Kai Zhang, Q. Kang, Le Pan, X. Wang, Can Cui","doi":"10.1109/ICNSC.2019.8743182","DOIUrl":"https://doi.org/10.1109/ICNSC.2019.8743182","url":null,"abstract":"One of the difficulties in computer vision is how to build an accurate classifier for a new target domain with insufficient labeled images from a related source domain with labeled images. Adversarial learning is a novel domain adaptation method that tackles this challenge by training robust deep networks and reducing the distribution difference between source and target domains, thus improving the classification performance on a target task. However, most prior adversarial adaptation learning approaches merely reduce the distribution difference across domains through GAN (Generative Adversarial Networks)-based loss, but when the performance of a generator or discriminator in GAN is degraded, the distribution difference between source and target domains are difficult to decrease. In this paper, we propose a novel generalized framework for adversarial domain adaptation, referred to as Generative Adversarial Distribution Matching. Our idea is to add the data discrepancy distance between source and target domains to the objective function of the generator so as to reduce distribution difference across domains through a Generator and a Discriminator compete against each other. Comprehensive experimental results confirm that it can well outperform several state-of-the-art methods for cross-domain image classification problems.","PeriodicalId":291695,"journal":{"name":"2019 IEEE 16th International Conference on Networking, Sensing and Control (ICNSC)","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128081550","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 : 2019-05-09DOI: 10.1109/ICNSC.2019.8743332
Hong Liu, Zizhen Zhang, Xiwang Guo
Vehicle routing problems (VRPs) are classical NP-hard problems. Those large scale and complex VRPs are even challenging. In this paper, we provide a general description of VRPs with heterogeneous constraints. For solving such type of VRPs in considerable solution quality and reasonable time, neighborhood search is a preferred choice. However, during the neighborhood search process, we may encounter a problem that the size of the neighboring solutions is still quite large, which consumes a great deal of computational resources. To tackle this problem, we propose a restricted neighborhood search method, which can ignore those non-promising neighboring solutions in a heuristic manner. Experiments on a real-world dataset show that our method can significantly accelerate the neighborhood search process, while the quality of the resultant solution is not impaired.
{"title":"Restricted Neighborhood Search for Large Scale Vehicle Routing Problems","authors":"Hong Liu, Zizhen Zhang, Xiwang Guo","doi":"10.1109/ICNSC.2019.8743332","DOIUrl":"https://doi.org/10.1109/ICNSC.2019.8743332","url":null,"abstract":"Vehicle routing problems (VRPs) are classical NP-hard problems. Those large scale and complex VRPs are even challenging. In this paper, we provide a general description of VRPs with heterogeneous constraints. For solving such type of VRPs in considerable solution quality and reasonable time, neighborhood search is a preferred choice. However, during the neighborhood search process, we may encounter a problem that the size of the neighboring solutions is still quite large, which consumes a great deal of computational resources. To tackle this problem, we propose a restricted neighborhood search method, which can ignore those non-promising neighboring solutions in a heuristic manner. Experiments on a real-world dataset show that our method can significantly accelerate the neighborhood search process, while the quality of the resultant solution is not impaired.","PeriodicalId":291695,"journal":{"name":"2019 IEEE 16th International Conference on Networking, Sensing and Control (ICNSC)","volume":"87 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133479363","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 studies a new single machine scheduling problem with sequence-dependent setup time, release time, due time and group technology assumption originated from a wire rod and bar rolling process in steel plants. The objective is to find an optimal batch sequence and job sequences of all batches to minimize the number of late jobs. A two-stage mixed integer program is created to describe and solve this problem. The first stage can be solved in a short time by CPLEX while the second one is time-consuming when dealing with large-scale cases. Thus, an iterated greedy algorithm able to solve the second stage fast is developed. The experimental results demonstrate that the proposed two-stage mixed integer program can be solved optimally by CPLEX for small-scale cases and the proposed algorithm can effectively solve the second stage for large-scale cases.
{"title":"Iterated Greedy Algorithm for Solving a New Single Machine Scheduling Problem","authors":"Ziyan Zhao, Shixin Liu, Mengchu Zhou, Xiwang Guo, JiaLun Xue","doi":"10.1109/ICNSC.2019.8743328","DOIUrl":"https://doi.org/10.1109/ICNSC.2019.8743328","url":null,"abstract":"This paper studies a new single machine scheduling problem with sequence-dependent setup time, release time, due time and group technology assumption originated from a wire rod and bar rolling process in steel plants. The objective is to find an optimal batch sequence and job sequences of all batches to minimize the number of late jobs. A two-stage mixed integer program is created to describe and solve this problem. The first stage can be solved in a short time by CPLEX while the second one is time-consuming when dealing with large-scale cases. Thus, an iterated greedy algorithm able to solve the second stage fast is developed. The experimental results demonstrate that the proposed two-stage mixed integer program can be solved optimally by CPLEX for small-scale cases and the proposed algorithm can effectively solve the second stage for large-scale cases.","PeriodicalId":291695,"journal":{"name":"2019 IEEE 16th International Conference on Networking, Sensing and Control (ICNSC)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130640025","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 : 2019-05-09DOI: 10.1109/ICNSC.2019.8743161
Xiwang Guo, Mengchu Zhou, Yaping Fu, Liang Qi, Dan You
In an actual remanufacturing process, a human operator is able to continuously learn the disassembly knowledge of an end-of-life product when he/she disassembles it, which makes him/her disassemble it more proficiently. In order to describe this feature, this work proposes a stochastic dual-objective disassembly sequencing planning problem considering human learning effects. In this problem, actual disassembly and setup time of operations are a function of their normal time and starting time. A new mathematical model is established to maximize total disassembly profit and minimize disassembly time. In order to solve this problem efficiently, a multi-population multi-objective evolutionary algorithm is developed. In this algorithm, some special strategies, e.g., solution representation, crossover operator and mutation operator, are newly designed based on this problem’s characteristics. Its effectiveness is well illustrated through several numerical cases and by comparing it with two prior approaches, i.e., nondominated sorting genetic algorithm II and multi-objective evolutionary algorithm based on decomposition. Experimental results demonstrate that the proposed algorithm performs well in solving this problem.
{"title":"Stochastic Dual-objective Disassembly Sequence Planning with Consideration of Learning Effect","authors":"Xiwang Guo, Mengchu Zhou, Yaping Fu, Liang Qi, Dan You","doi":"10.1109/ICNSC.2019.8743161","DOIUrl":"https://doi.org/10.1109/ICNSC.2019.8743161","url":null,"abstract":"In an actual remanufacturing process, a human operator is able to continuously learn the disassembly knowledge of an end-of-life product when he/she disassembles it, which makes him/her disassemble it more proficiently. In order to describe this feature, this work proposes a stochastic dual-objective disassembly sequencing planning problem considering human learning effects. In this problem, actual disassembly and setup time of operations are a function of their normal time and starting time. A new mathematical model is established to maximize total disassembly profit and minimize disassembly time. In order to solve this problem efficiently, a multi-population multi-objective evolutionary algorithm is developed. In this algorithm, some special strategies, e.g., solution representation, crossover operator and mutation operator, are newly designed based on this problem’s characteristics. Its effectiveness is well illustrated through several numerical cases and by comparing it with two prior approaches, i.e., nondominated sorting genetic algorithm II and multi-objective evolutionary algorithm based on decomposition. Experimental results demonstrate that the proposed algorithm performs well in solving this problem.","PeriodicalId":291695,"journal":{"name":"2019 IEEE 16th International Conference on Networking, Sensing and Control (ICNSC)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128692893","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 : 2019-05-09DOI: 10.1109/ICNSC.2019.8743211
Haixu Yu, Pei Yang
Reward is one of the crucial factors in reinforcement learning, which affects the improvement of control strategies. However, the role of reward design has received relatively little attention. In this paper, an emotion-based target reward function is proposed which requires the agent to possess the ability to reflect. In this approach, the learning process information of the agent is mapped to its internal changes in any episode. The difference in internal values of adjacent episodes induces the generation of the agent’s emotions, which is a key way to assist the agent to internally measure preset external target reward. Our proposed approach is combined with traditional RL algorithms (i.e., Q-learning, Sarsa and Q($lambda$)-learning) to test its effectiveness. All experimental results show that emotion-based target reward can accelerate the learning process.
{"title":"An Emotion-Based Approach to Reinforcement Learning Reward Design","authors":"Haixu Yu, Pei Yang","doi":"10.1109/ICNSC.2019.8743211","DOIUrl":"https://doi.org/10.1109/ICNSC.2019.8743211","url":null,"abstract":"Reward is one of the crucial factors in reinforcement learning, which affects the improvement of control strategies. However, the role of reward design has received relatively little attention. In this paper, an emotion-based target reward function is proposed which requires the agent to possess the ability to reflect. In this approach, the learning process information of the agent is mapped to its internal changes in any episode. The difference in internal values of adjacent episodes induces the generation of the agent’s emotions, which is a key way to assist the agent to internally measure preset external target reward. Our proposed approach is combined with traditional RL algorithms (i.e., Q-learning, Sarsa and Q($lambda$)-learning) to test its effectiveness. All experimental results show that emotion-based target reward can accelerate the learning process.","PeriodicalId":291695,"journal":{"name":"2019 IEEE 16th International Conference on Networking, Sensing and Control (ICNSC)","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128939609","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 : 2019-05-09DOI: 10.1109/ICNSC.2019.8743164
Anthony Thomas, Yunhui Guo, Yeseong Kim, Baris Aksanli, Arun Kumar, T. Simunic
Networked applications with heterogeneous sensors are a growing source of data. Such applications use machine learning (ML) to make real-time predictions. Currently, features from all sensors are collected in a centralized cloud-based tier to form the whole feature vector for ML prediction. This approach has high communication cost, which wastes energy and often bottlenecks the network. In this work, we study an alternative approach that mitigates such issues by “pushing” ML inference computations out of the cloud and onto a hierarchy of IoT devices. Our approach presents a new technical challenge of “rewriting” an ML inference computation to factor it over a network of devices without significantly reducing prediction accuracy. We introduce novel exact factoring algorithms for some popular models that preserve accuracy. We also create novel approximate variants of other models that offer high accuracy. Measurements on a common IoT device show that energy use and latency can be reduced by up to 63% and 67% respectively without reducing accuracy relative to sending all data to the cloud.
{"title":"Hierarchical and Distributed Machine Learning Inference Beyond the Edge","authors":"Anthony Thomas, Yunhui Guo, Yeseong Kim, Baris Aksanli, Arun Kumar, T. Simunic","doi":"10.1109/ICNSC.2019.8743164","DOIUrl":"https://doi.org/10.1109/ICNSC.2019.8743164","url":null,"abstract":"Networked applications with heterogeneous sensors are a growing source of data. Such applications use machine learning (ML) to make real-time predictions. Currently, features from all sensors are collected in a centralized cloud-based tier to form the whole feature vector for ML prediction. This approach has high communication cost, which wastes energy and often bottlenecks the network. In this work, we study an alternative approach that mitigates such issues by “pushing” ML inference computations out of the cloud and onto a hierarchy of IoT devices. Our approach presents a new technical challenge of “rewriting” an ML inference computation to factor it over a network of devices without significantly reducing prediction accuracy. We introduce novel exact factoring algorithms for some popular models that preserve accuracy. We also create novel approximate variants of other models that offer high accuracy. Measurements on a common IoT device show that energy use and latency can be reduced by up to 63% and 67% respectively without reducing accuracy relative to sending all data to the cloud.","PeriodicalId":291695,"journal":{"name":"2019 IEEE 16th International Conference on Networking, Sensing and Control (ICNSC)","volume":"34 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-05-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115938655","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：