Pub Date : 2020-04-17DOI: 10.1504/ijcse.2020.10028617
I. Grigoryev, A. Bagirov, M. Tuck
Inaccurate drill-core assay interpretation in the exploration stage presents challenges to long-term profit of gold mining operations. Predicting the gold distribution within a deposit as precisely as possible is one of the most important aspects of the methodologies employed to avoid problems associated with financial expectations. The prediction of the variability of gold using a very limited number of drill-core samples is a very challenging problem. This is often intractable using traditional statistical tools where with less than complete spatial information certain assumptions are made about gold distribution and mineralisation. The decision-support predictive modelling methodology based on the unsupervised machine learning technique, presented in this paper avoids some of the restrictive limitations of traditional methods. It identifies promising exploration targets missed during exploration and recovers hidden spatial and physical characteristics of the explored deposit using information directly from drill hole database.
{"title":"Prediction of gold-bearing localised occurrences from limited exploration data","authors":"I. Grigoryev, A. Bagirov, M. Tuck","doi":"10.1504/ijcse.2020.10028617","DOIUrl":"https://doi.org/10.1504/ijcse.2020.10028617","url":null,"abstract":"Inaccurate drill-core assay interpretation in the exploration stage presents challenges to long-term profit of gold mining operations. Predicting the gold distribution within a deposit as precisely as possible is one of the most important aspects of the methodologies employed to avoid problems associated with financial expectations. The prediction of the variability of gold using a very limited number of drill-core samples is a very challenging problem. This is often intractable using traditional statistical tools where with less than complete spatial information certain assumptions are made about gold distribution and mineralisation. The decision-support predictive modelling methodology based on the unsupervised machine learning technique, presented in this paper avoids some of the restrictive limitations of traditional methods. It identifies promising exploration targets missed during exploration and recovers hidden spatial and physical characteristics of the explored deposit using information directly from drill hole database.","PeriodicalId":340410,"journal":{"name":"Int. J. Comput. Sci. Eng.","volume":"62 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131211412","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 : 2020-04-17DOI: 10.1504/ijcse.2020.10028625
Guto Leoni Santos, D. Gomes, J. Kelner, D. Sadok, Francisco Airton Silva, P. Endo, Theo Lynn
E-health systems can be used to monitor people in real-time, offering a range of multimedia-based health services, at the same time reducing the cost since cheaper devices can be used to compose it. However, any downtime, mainly in the case of critical health services, can result in patient health problems and in the worst case, loss of life. In this paper, we use an interdisciplinary approach combining stochastic models with optimisation algorithms to analyse how failures impact e-health monitoring system availability. We propose surrogate models to estimate the availability of e-health monitoring systems that rely on edge, fog, and cloud infrastructures. Then, we apply a multi-objective optimisation algorithm, NSGA-II, to improve system availability considering component costs as constraint. Results suggest that replacing components with more reliable ones is more effective in improving the availability of an e-health monitoring system than adding more redundant components.
{"title":"The internet of things for healthcare: optimising e-health system availability in the fog and cloud","authors":"Guto Leoni Santos, D. Gomes, J. Kelner, D. Sadok, Francisco Airton Silva, P. Endo, Theo Lynn","doi":"10.1504/ijcse.2020.10028625","DOIUrl":"https://doi.org/10.1504/ijcse.2020.10028625","url":null,"abstract":"E-health systems can be used to monitor people in real-time, offering a range of multimedia-based health services, at the same time reducing the cost since cheaper devices can be used to compose it. However, any downtime, mainly in the case of critical health services, can result in patient health problems and in the worst case, loss of life. In this paper, we use an interdisciplinary approach combining stochastic models with optimisation algorithms to analyse how failures impact e-health monitoring system availability. We propose surrogate models to estimate the availability of e-health monitoring systems that rely on edge, fog, and cloud infrastructures. Then, we apply a multi-objective optimisation algorithm, NSGA-II, to improve system availability considering component costs as constraint. Results suggest that replacing components with more reliable ones is more effective in improving the availability of an e-health monitoring system than adding more redundant components.","PeriodicalId":340410,"journal":{"name":"Int. J. Comput. Sci. Eng.","volume":"321 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121401366","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}
Sentence semantic matching (SSM) is a fundamental research task in natural language processing. Most existing SSM methods take the advantage of sentence representation learning to generate a single or multi-granularity semantic representation for sentence matching. However, sentence interactions and loss function which are the two key factors for SSM still have not been fully considered. Accordingly, we propose a deep neural network architecture for SSM task with a sentence interactive matching layer and an optimised loss function. Given two input sentences, our model first encodes them to embeddings with an ordinary long short-term memory (LSTM) encoder. Then, the encoded embeddings are handled by an attention layer to find the key and important words in the sentences. Next, sentence interactions are captured with a matching layer to output a matching vector. Finally, based on the matching vector, a fully connected multi-layer perceptron outputs the similarity score. The model also distinguishes the equivocation training instances with an improved optimised loss function. We also systematically evaluate our model on a public Chinese semantic matching corpus, BQ corpus. The results demonstrate that our model outperforms the state-of-the-art methods, i.e., BiMPM, DIIN.
{"title":"A deep neural architecture for sentence semantic matching","authors":"Xu Zhang, Wenpeng Lu, Fangfang Li, Ruoyu Zhang, Jinyong Cheng","doi":"10.1504/ijcse.2020.10028622","DOIUrl":"https://doi.org/10.1504/ijcse.2020.10028622","url":null,"abstract":"Sentence semantic matching (SSM) is a fundamental research task in natural language processing. Most existing SSM methods take the advantage of sentence representation learning to generate a single or multi-granularity semantic representation for sentence matching. However, sentence interactions and loss function which are the two key factors for SSM still have not been fully considered. Accordingly, we propose a deep neural network architecture for SSM task with a sentence interactive matching layer and an optimised loss function. Given two input sentences, our model first encodes them to embeddings with an ordinary long short-term memory (LSTM) encoder. Then, the encoded embeddings are handled by an attention layer to find the key and important words in the sentences. Next, sentence interactions are captured with a matching layer to output a matching vector. Finally, based on the matching vector, a fully connected multi-layer perceptron outputs the similarity score. The model also distinguishes the equivocation training instances with an improved optimised loss function. We also systematically evaluate our model on a public Chinese semantic matching corpus, BQ corpus. The results demonstrate that our model outperforms the state-of-the-art methods, i.e., BiMPM, DIIN.","PeriodicalId":340410,"journal":{"name":"Int. J. Comput. Sci. Eng.","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125064803","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 : 2020-04-17DOI: 10.1504/ijcse.2020.10028620
Randa Benkhelifa, Ismaïl Biskri, F. Z. Laallam, Esma Aïmeur
Social networking websites are growing not only regarding the number of users but also in terms of the user-generated content. These data represent a valuable source of information for several applications, which require the meaning of that content associated with the personal data. However, the current structure of social networks does not allow extracting in a fast and straightforward way the hidden information sought by these applications. Major efforts have emerged from the semantic web community addressing this problem trying to represent the user as accurately as possible. They are not unable to give a sense to the user-generated content. For this, more sense-making needs to be done on the content, to enrich the user profile. In this paper, we introduce a generic model called user content categorisation (UCC). It incorporates the text mining approach into a semantic model to enrich the user profile by including information on user's posts classifications.
{"title":"User content categorisation model, a generic model that combines text mining and semantic models","authors":"Randa Benkhelifa, Ismaïl Biskri, F. Z. Laallam, Esma Aïmeur","doi":"10.1504/ijcse.2020.10028620","DOIUrl":"https://doi.org/10.1504/ijcse.2020.10028620","url":null,"abstract":"Social networking websites are growing not only regarding the number of users but also in terms of the user-generated content. These data represent a valuable source of information for several applications, which require the meaning of that content associated with the personal data. However, the current structure of social networks does not allow extracting in a fast and straightforward way the hidden information sought by these applications. Major efforts have emerged from the semantic web community addressing this problem trying to represent the user as accurately as possible. They are not unable to give a sense to the user-generated content. For this, more sense-making needs to be done on the content, to enrich the user profile. In this paper, we introduce a generic model called user content categorisation (UCC). It incorporates the text mining approach into a semantic model to enrich the user profile by including information on user's posts classifications.","PeriodicalId":340410,"journal":{"name":"Int. J. Comput. Sci. Eng.","volume":"51 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122845694","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 : 2020-04-17DOI: 10.1504/ijcse.2020.10028619
Edna Chebet Too, Li Yujian, P. K. Gadosey, Sam Njuki, Firdaous Essaf
Deep learning architectures which are exceptionally deep have exhibited to be incredibly powerful models for image processing. As the architectures become deep, it introduces challenges and difficulties in the training process such as overfitting, computational cost, and exploding/vanishing gradients and degradation. A new state-of-the-art densely connected architecture, called DenseNets, has exhibited an exceptionally outstanding result for image classification. However, it still computationally costly to train DenseNets. The choice of the activation function is also an important aspect in training of deep learning networks because it has a considerable impact on the training and performance of a network model. Therefore, an empirical analysis of some of the nonlinear activation functions used in deep learning is done for image classification. The activation functions evaluated include ReLU, Leaky ReLU, ELU, SELU and an ensemble of SELU and ELU. Publicly available datasets Cifar-10, SVHN, and PlantVillage are used for evaluation.
{"title":"Performance analysis of nonlinear activation function in convolution neural network for image classification","authors":"Edna Chebet Too, Li Yujian, P. K. Gadosey, Sam Njuki, Firdaous Essaf","doi":"10.1504/ijcse.2020.10028619","DOIUrl":"https://doi.org/10.1504/ijcse.2020.10028619","url":null,"abstract":"Deep learning architectures which are exceptionally deep have exhibited to be incredibly powerful models for image processing. As the architectures become deep, it introduces challenges and difficulties in the training process such as overfitting, computational cost, and exploding/vanishing gradients and degradation. A new state-of-the-art densely connected architecture, called DenseNets, has exhibited an exceptionally outstanding result for image classification. However, it still computationally costly to train DenseNets. The choice of the activation function is also an important aspect in training of deep learning networks because it has a considerable impact on the training and performance of a network model. Therefore, an empirical analysis of some of the nonlinear activation functions used in deep learning is done for image classification. The activation functions evaluated include ReLU, Leaky ReLU, ELU, SELU and an ensemble of SELU and ELU. Publicly available datasets Cifar-10, SVHN, and PlantVillage are used for evaluation.","PeriodicalId":340410,"journal":{"name":"Int. J. Comput. Sci. Eng.","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122124387","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 : 2020-03-13DOI: 10.1504/ijcse.2020.10027622
Fethallah Hadjila, Amine Belabed, M. Merzoug
The QoS-based service selection in a highly dynamical environment is becoming a challenging issue. In practice, the QoS fluctuations of a service composition entail major difficulties in measuring the degree to which the user requirements are satisfied. In addition, the search space of feasible compositions (i.e., the solutions that preserve the requirements) is generally large and cannot be explored in a limited time; therefore, we need an approach that not only copes with the presence of uncertainty but also ensures a pertinent search with a reduced computational cost. To tackle this problem, we propose a constraint programming framework and a set of ranking heuristics that both reduce the search space and ensure a set of reliable compositions. The conducted experiments show that the ranking heuristics, termed 'fuzzy dominance' and 'probabilistic skyline', outperform almost all existing state-of-the-art methods.
{"title":"Efficient web service selection with uncertain QoS","authors":"Fethallah Hadjila, Amine Belabed, M. Merzoug","doi":"10.1504/ijcse.2020.10027622","DOIUrl":"https://doi.org/10.1504/ijcse.2020.10027622","url":null,"abstract":"The QoS-based service selection in a highly dynamical environment is becoming a challenging issue. In practice, the QoS fluctuations of a service composition entail major difficulties in measuring the degree to which the user requirements are satisfied. In addition, the search space of feasible compositions (i.e., the solutions that preserve the requirements) is generally large and cannot be explored in a limited time; therefore, we need an approach that not only copes with the presence of uncertainty but also ensures a pertinent search with a reduced computational cost. To tackle this problem, we propose a constraint programming framework and a set of ranking heuristics that both reduce the search space and ensure a set of reliable compositions. The conducted experiments show that the ranking heuristics, termed 'fuzzy dominance' and 'probabilistic skyline', outperform almost all existing state-of-the-art methods.","PeriodicalId":340410,"journal":{"name":"Int. J. Comput. Sci. Eng.","volume":"26 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124369042","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 : 2020-03-13DOI: 10.1504/ijcse.2020.10027621
Naveen Kumar, Manoj Agarwal, S. Deshmukh, Shikha Gupta
Process mining aims at discovering the workflow of a process from the event logs that provide insights into organisational processes for improving these processes and their support systems. Process mining abstracts the complex real-life datasets into a well-structured form known as a process model. In an ideal scenario, a process mining algorithm should produce a model that is simple, precise, general and fits the available logs. A conventional process mining algorithm typically generates a single process model that may not describe the recorded behaviour effectively. Multi-objective evolutionary algorithms (MOEA) for process mining optimise two or more objectives to generate several competing process models from the event logs. Subsequently, a user can choose a model based on his/her preference. In this paper, we have experimentally compared the popular second-generation MOEA algorithms for process mining.
{"title":"MOEA for discovering Pareto-optimal process models: an experimental comparison","authors":"Naveen Kumar, Manoj Agarwal, S. Deshmukh, Shikha Gupta","doi":"10.1504/ijcse.2020.10027621","DOIUrl":"https://doi.org/10.1504/ijcse.2020.10027621","url":null,"abstract":"Process mining aims at discovering the workflow of a process from the event logs that provide insights into organisational processes for improving these processes and their support systems. Process mining abstracts the complex real-life datasets into a well-structured form known as a process model. In an ideal scenario, a process mining algorithm should produce a model that is simple, precise, general and fits the available logs. A conventional process mining algorithm typically generates a single process model that may not describe the recorded behaviour effectively. Multi-objective evolutionary algorithms (MOEA) for process mining optimise two or more objectives to generate several competing process models from the event logs. Subsequently, a user can choose a model based on his/her preference. In this paper, we have experimentally compared the popular second-generation MOEA algorithms for process mining.","PeriodicalId":340410,"journal":{"name":"Int. J. Comput. Sci. Eng.","volume":"26 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126048259","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 : 2020-03-13DOI: 10.1504/ijcse.2020.10027620
Li Li, Yifei Wei, Mei Song, Xiaojun Wang
Software defined networking (SDN) is revolutionising the telecommunication networking industry by providing flexible and efficient management. This paper proposes an energy-efficiency-aware flow-based management framework for relay-assisted heterogeneous networks (HetNets), where the relay nodes are powered by renewable energy. Due to the dynamic property of user behaviour and renewable energy availability, the flow-based management layer should enhance not only the instantaneous energy efficiency but also the long-term energy efficiency, while satisfying the transmission rate demand for each user. We first formulate the energy efficiency problem in HetNets as an optimisation problem for instantaneous energy efficiency and renewable energy allocation, and propose a heuristic algorithm to solve the optimisation problem. According to the proposed algorithm, we then design a dynamic flow-table configuration policy (DFTCP) which can be integrated as an application on top of an SDN controller to enhance the long-term energy efficiency. Simulation results show that the proposed policy can achieve higher energy efficiency compared with current distributed relay strategy, which chooses the nearest or strongest signal node to access, and obtain better performance for the overall relay network when the user density and demand change.
{"title":"Energy-efficiency-aware flow-based access control in HetNets with renewable energy supply","authors":"Li Li, Yifei Wei, Mei Song, Xiaojun Wang","doi":"10.1504/ijcse.2020.10027620","DOIUrl":"https://doi.org/10.1504/ijcse.2020.10027620","url":null,"abstract":"Software defined networking (SDN) is revolutionising the telecommunication networking industry by providing flexible and efficient management. This paper proposes an energy-efficiency-aware flow-based management framework for relay-assisted heterogeneous networks (HetNets), where the relay nodes are powered by renewable energy. Due to the dynamic property of user behaviour and renewable energy availability, the flow-based management layer should enhance not only the instantaneous energy efficiency but also the long-term energy efficiency, while satisfying the transmission rate demand for each user. We first formulate the energy efficiency problem in HetNets as an optimisation problem for instantaneous energy efficiency and renewable energy allocation, and propose a heuristic algorithm to solve the optimisation problem. According to the proposed algorithm, we then design a dynamic flow-table configuration policy (DFTCP) which can be integrated as an application on top of an SDN controller to enhance the long-term energy efficiency. Simulation results show that the proposed policy can achieve higher energy efficiency compared with current distributed relay strategy, which chooses the nearest or strongest signal node to access, and obtain better performance for the overall relay network when the user density and demand change.","PeriodicalId":340410,"journal":{"name":"Int. J. Comput. Sci. Eng.","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131222816","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 : 2020-03-13DOI: 10.1504/ijcse.2020.10027617
Mingming Wang, Zhiguo Qu, Lin-Ming Gong
With the rapid progress of quantum cryptography, secret sharing has been developed in the quantum setting for achieving a high level of security, which is known as quantum secret sharing (QSS). The first QSS scheme was proposed by Hillery et al. in 1999 [Phys. Rev. A, Vol. 59, p.1829 (1999)] based on entangled Greenberger-Horne-Zeilinger (GHZ) states. However, only 50% of the entangled quantum states are effective for eavesdropping detection and secret splitting in the original scheme. In this paper, we introduce a possible method, called measurement-delay strategy, to improve the qubit efficiency of the GHZ-based QSS scheme. By using this method, the qubit efficiency of the improved QSS scheme can reach 100% for both security detection and secret distribution. The improved QSS scheme can be implemented experimentally based on current technologies.
{"title":"Improved quantum secret sharing scheme based on GHZ states","authors":"Mingming Wang, Zhiguo Qu, Lin-Ming Gong","doi":"10.1504/ijcse.2020.10027617","DOIUrl":"https://doi.org/10.1504/ijcse.2020.10027617","url":null,"abstract":"With the rapid progress of quantum cryptography, secret sharing has been developed in the quantum setting for achieving a high level of security, which is known as quantum secret sharing (QSS). The first QSS scheme was proposed by Hillery et al. in 1999 [Phys. Rev. A, Vol. 59, p.1829 (1999)] based on entangled Greenberger-Horne-Zeilinger (GHZ) states. However, only 50% of the entangled quantum states are effective for eavesdropping detection and secret splitting in the original scheme. In this paper, we introduce a possible method, called measurement-delay strategy, to improve the qubit efficiency of the GHZ-based QSS scheme. By using this method, the qubit efficiency of the improved QSS scheme can reach 100% for both security detection and secret distribution. The improved QSS scheme can be implemented experimentally based on current technologies.","PeriodicalId":340410,"journal":{"name":"Int. J. Comput. Sci. Eng.","volume":"39 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126769683","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 : 2020-03-13DOI: 10.1504/ijcse.2020.10027619
Zikai Nie, Zhisheng Li, Lei Wang, Shasha Guo, Yu Deng, Rangyu Deng, Q. Dou
With the development of convolutional neural networks (CNNs), their high computational complexity and energy consumption become significant problems. Many CNN inference accelerators are proposed to reduce the consumption. Most of them are based on 32-bit float-point matrix multiplication, where the data precision is over-provisioned. This paper presents Laius, an 8-bit fixed-point LeNet inference engine implemented on FPGA. To achieve low-precision computation and storage, we introduce our fixed-point training framework called FixCaffe. To economise FPGA resources, we proposed a methodology to find the optimal bit-length for weight and bias in LeNet. We use optimisations of pipelining, tiling, and theoretical analysis to improve the performance. Experiment results show that Laius achieves 44.9 Gops throughputs. Moreover, with only 1% accuracy loss, 8-bit Laius largely reduces 31.43% in delay, 87.01% in LUT consumption, 66.50% in BRAM consumption, 65.11% in DSP consumption and 47.95% in power compared to the 32-bit version with the same structure.
{"title":"Laius: an energy-efficient FPGA CNN accelerator with the support of a fixed-point training framework","authors":"Zikai Nie, Zhisheng Li, Lei Wang, Shasha Guo, Yu Deng, Rangyu Deng, Q. Dou","doi":"10.1504/ijcse.2020.10027619","DOIUrl":"https://doi.org/10.1504/ijcse.2020.10027619","url":null,"abstract":"With the development of convolutional neural networks (CNNs), their high computational complexity and energy consumption become significant problems. Many CNN inference accelerators are proposed to reduce the consumption. Most of them are based on 32-bit float-point matrix multiplication, where the data precision is over-provisioned. This paper presents Laius, an 8-bit fixed-point LeNet inference engine implemented on FPGA. To achieve low-precision computation and storage, we introduce our fixed-point training framework called FixCaffe. To economise FPGA resources, we proposed a methodology to find the optimal bit-length for weight and bias in LeNet. We use optimisations of pipelining, tiling, and theoretical analysis to improve the performance. Experiment results show that Laius achieves 44.9 Gops throughputs. Moreover, with only 1% accuracy loss, 8-bit Laius largely reduces 31.43% in delay, 87.01% in LUT consumption, 66.50% in BRAM consumption, 65.11% in DSP consumption and 47.95% in power compared to the 32-bit version with the same structure.","PeriodicalId":340410,"journal":{"name":"Int. J. Comput. Sci. Eng.","volume":"83 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132823316","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：