論文 - 田中 剛平
-
Keynote Speech: Information processing hardware, physical reservoir computing and complex-valued neural networks
Akira Hirose, Ryosho Nakane, Gouhei Tanaka
IMFEDK 2019 - International Meeting for Future of Electron Devices, Kansai 19 - 24 2019年11月
記述言語：英語 掲載種別：研究論文（国際会議プロシーディングス） 出版者・発行元：Institute of Electrical and Electronics Engineers Inc.
First we discuss the essence of neural networks, which are the bases of modern artificial intelligence (AI), to examine the relationship between the neural fundamental framework and the present hardware. Then, in this context, we review reservoir computing and complex-valued neural networks.
-
Intervention threshold for epidemic control in susceptible-infected-recovered metapopulation models. 査読あり 国際誌
Akari Matsuki, Gouhei Tanaka
Physical review. E 100 ( 2-1 ) 022302 - 022302 2019年08月
担当区分：最終著者 記述言語：英語 掲載種別：研究論文（学術雑誌）
Metapopulation epidemic models describe epidemic dynamics in networks of spatially distant patches connected via pathways for migration of individuals. In the present study, we deal with a susceptible-infected-recovered (SIR) metapopulation model where the epidemic process in each patch is represented by an SIR model and the mobility of individuals is assumed to be a homogeneous diffusion. We consider two types of patches including high-risk and low-risk ones under the assumption that a local patch is changed from a high-risk one to a low-risk one by an intervention. We theoretically analyze the intervention threshold which indicates the critical fraction of low-risk patches for preventing a global epidemic outbreak. We show that an intervention targeted to high-degree patches is more effective for epidemic control than a random intervention. The theoretical results are validated by Monte Carlo simulations for synthetic and realistic scale-free patch networks. The theoretical results also reveal that the intervention threshold depends on the human mobility network and the mobility rate. Our approach is useful for exploring better local interventions aimed at containment of epidemics.
-
Hybrid pooling for enhancement of generalization ability in deep convolutional neural networks 査読あり
Zhiqiang Tong, Gouhei Tanaka
NEUROCOMPUTING 333 76 - 85 2019年03月
担当区分：最終著者, 責任著者 記述言語：英語 掲載種別：研究論文（学術雑誌） 出版者・発行元：ELSEVIER
Convolutional neural networks (CNNs) have attracted considerable attention in many application fields for their great ability to deal with image recognition and object detection tasks. A pooling process is an important process in CNNs, which serves to decrease the dimensionality of processed data for reducing computational cost as well as for enhancing tolerance to translation and noise. Although standard pooling methods, such as the max pooling and the average pooling, are typically adopted in many studies, a newly devised pooling method could improve the generalization ability of CNNs. In this study, we propose a hybrid pooling method which stochastically chooses the max pooling or the average pooling in each pooling layer. A characteristic of the hybrid pooling is that the probability for choosing one of the two pooling methods can be controlled for each convolutional layer. In image classification tasks with benchmark datasets, we show that the hybrid pooling is effective for increasing the generalization ability of CNNs. Moreover, we demonstrate that the hybrid pooling combined with the dropout is competitive with other existing methods in classification performance. (C) 2018 Elsevier B.V. All rights reserved.
-
Analysis on Characteristics of Multi-Step Learning Echo State Networks for Nonlinear Time Series Prediction. 査読あり
Takanori Akiyama, Gouhei Tanaka
International Joint Conference on Neural Networks(IJCNN) 1 - 8 2019年
担当区分：最終著者 掲載種別：研究論文（国際会議プロシーディングス） 出版者・発行元：IEEE
DOI： 10.1109/IJCNN.2019.8851876
その他リンク： https://dblp.uni-trier.de/db/conf/ijcnn/ijcnn2019.html#AkiyamaT19
-
In a Spin-Wave Reservoir for Machine Learning. 査読あり
Ryosho Nakane, Gouhei Tanaka, Akira Hirose
International Joint Conference on Neural Networks(IJCNN) 1 - 9 2019年
掲載種別：研究論文（国際会議プロシーディングス） 出版者・発行元：IEEE
DOI： 10.1109/IJCNN.2019.8852280
その他リンク： https://dblp.uni-trier.de/db/conf/ijcnn/ijcnn2019.html#NakaneTH19
-
Echo State Networks Composed of Units with Time-Varying Nonlinearity. 査読あり
Gouhei Tanaka, Ryosho Nakane, Akira Hirose
Aust. J. Intell. Inf. Process. Syst. 17 ( 2 ) 34 - 39 2019年
担当区分：筆頭著者 掲載種別：研究論文（学術雑誌）
その他リンク： https://dblp.uni-trier.de/db/journals/ajiips/ajiips17.html#TanakaNH19
-
Echo State Network with Adversarial Training. 査読あり
Takanori Akiyama, Gouhei Tanaka
Artificial Neural Networks and Machine Learning - ICANN 2019 - 28th International Conference on Artificial Neural Networks, Munich, Germany, September 17-19, 2019, Proceedings - Workshop and Special Sessions 82 - 88 2019年
担当区分：最終著者 掲載種別：研究論文（国際会議プロシーディングス） 出版者・発行元：Springer
DOI： 10.1007/978-3-030-30493-5_8
その他リンク： https://dblp.uni-trier.de/db/conf/icann/icann2019w.html#AkiyamaT19
-
Application Identification of Network Traffic by Reservoir Computing 査読あり
Toshiyuki Yamane, Jean Benoit Heroux, Hidetoshi Numata, Gouhei Tanaka, Ryosho Nakane, Akira Hirose
NEURAL INFORMATION PROCESSING, ICONIP 2019, PT V 1143 389 - 396 2019年
記述言語：英語 掲載種別：研究論文（国際会議プロシーディングス） 出版者・発行元：SPRINGER INTERNATIONAL PUBLISHING AG
We propose a method for application identification for network traffic by reservoir computing. Different from conventional approaches, the proposed method handles traffic flows as dynamical time series data and enables fast and real-time identification. We apply the proposed method to real traffic data and show that high identification accuracy is achieved. We also discuss an implementation as physical reservoirs based on optics and the impact of the proposed method to 5G networking.
DOI： 10.1007/978-3-030-36802-9_41
その他リンク： https://dblp.uni-trier.de/db/conf/iconip/iconip2019-5.html#YamaneHNTNH19
-
Physical reservoir computing: Possibility to resolve the inconsistency between neuro-AI principles and its hardware. 査読あり
Akira Hirose, Seiji Takeda, Toshiyuki Yamane, Hidetoshi Numata, Naoki Kanazawa, Jean Benoit Héroux, Daiju Nakano, Ryosho Nakane, Gouhei Tanaka
Aust. J. Intell. Inf. Process. Syst. 16 ( 4 ) 49 - 55 2019年
担当区分：最終著者 掲載種別：研究論文（学術雑誌）
その他リンク： https://dblp.uni-trier.de/db/journals/ajiips/ajiips16.html#HiroseTYNKHNNT19
-
Bifurcation mechanism for emergence of spontaneous oscillations in coupled heterogeneous excitable units 査読あり
Morino Kai, Tanaka Gouhei, Aihara Kazuyuki
PHYSICAL REVIEW E 98 ( 5 ) 2018年11月
-
Recent Advances in Physical Reservoir Computing: A Review. 査読あり
Gouhei Tanaka, Toshiyuki Yamane, Jean Benoit, Héroux, Ryosho Nakane, Naoki Kanazawa, Seiji Takeda, Hidetoshi Numata, Daiju Nakano, Akira Hirose
CoRR abs/1808.04962 2018年08月
掲載種別：研究論文（学術雑誌）
Reservoir computing is a computational framework suited for<br />
temporal/sequential data processing. It is derived from several recurrent<br />
neural network models, including echo state networks and liquid state machines.<br />
A reservoir computing system consists of a reservoir for mapping inputs into a<br />
high-dimensional space and a readout for extracting features of the inputs.<br />
Further, training is carried out only in the readout. Thus, the major advantage<br />
of reservoir computing is fast and simple learning compared to other recurrent<br />
neural networks. Another advantage is that the reservoir can be realized using<br />
physical systems, substrates, and devices, instead of recurrent neural<br />
networks. In fact, such physical reservoir computing has attracted increasing<br />
attention in various fields of research. The purpose of this review is to<br />
provide an overview of recent advances in physical reservoir computing by<br />
classifying them according to the type of the reservoir. We discuss the current<br />
issues and perspectives related to physical reservoir computing, in order to<br />
further expand its practical applications and develop next-generation machine<br />
learning systems. -
Bifurcation analysis of a mathematical model of atopic dermatitis to determine patient-specific effects of treatments on dynamic phenotypes. 査読あり 国際誌
Gouhei Tanaka, Elisa Domínguez-Hüttinger, Panayiotis Christodoulides, Kazuyuki Aihara, Reiko J Tanaka
Journal of theoretical biology 448 66 - 79 2018年07月
担当区分：筆頭著者, 責任著者 記述言語：英語 掲載種別：研究論文（学術雑誌） 出版者・発行元：Academic Press
Atopic dermatitis (AD) is a common inflammatory skin disease, whose incidence is currently increasing worldwide. AD has a complex etiology, involving genetic, environmental, immunological, and epidermal factors, and its pathogenic mechanisms have not yet been fully elucidated. Identification of AD risk factors and systematic understanding of their interactions are required for exploring effective prevention and treatment strategies for AD. We recently developed a mathematical model for AD pathogenesis to clarify mechanisms underlying AD onset and progression. This model describes a dynamic interplay between skin barrier, immune regulation, and environmental stress, and reproduced four types of dynamic behaviour typically observed in AD patients in response to environmental triggers. Here, we analyse bifurcations of the model to identify mathematical conditions for the system to demonstrate transitions between different types of dynamic behaviour that reflect respective severity of AD symptoms. By mathematically modelling effects of topical application of antibiotics, emollients, corticosteroids, and their combinations with different application schedules and doses, bifurcation analysis allows us to mathematically evaluate effects of the treatments on improving AD symptoms in terms of the patients' dynamic behaviour. The mathematical method developed in this study can be used to explore and improve patient-specific personalised treatment strategies to control AD symptoms.
-
Dimensionality Reduction by Reservoir Computing and Its Application to IoT Edge Computing 査読あり
Toshiyuki Yamane, Hidetoshi Numata, Jean Benoit Heroux, Naoki Kanazawa, Seiji Takeda, Gouhei Tanaka, Ryosho Nakane, Akira Hirose, Daiju Nakano
NEURAL INFORMATION PROCESSING (ICONIP 2018), PT I 11301 635 - 643 2018年
記述言語：英語 掲載種別：研究論文（国際会議プロシーディングス） 出版者・発行元：SPRINGER INTERNATIONAL PUBLISHING AG
We propose a method of dimension reduction of high dimensional time series data by reservoir computing. The proposed method is a generalization of random projection techniques to time series, which uses a reservoir smaller than input time series. We demonstrate the method by echo state networks for artificially generated time series data. We also discuss an implementation as physical reservoirs and its application of the proposed method to IoT edge computing, which is the first proposal for industry application of physical reservoir computing beyond standard benchmark tasks.
DOI： 10.1007/978-3-030-04167-0_58
その他リンク： https://dblp.uni-trier.de/db/conf/iconip/iconip2018-1.html#YamaneNHKTTNHN18
-
Prediction of Molecular Packing Motifs in Organic Crystals by Neural Graph Fingerprints. 査読あり
Daiki Ito, Raku Shirasawa, Shinnosuke Hattori, Shigetaka Tomiya, Gouhei Tanaka
Neural Information Processing - 25th International Conference, ICONIP 2018, Siem Reap, Cambodia, December 13-16, 2018, Proceedings, Part V 26 - 34 2018年
担当区分：最終著者, 責任著者 掲載種別：研究論文（国際会議プロシーディングス） 出版者・発行元：Springer
DOI： 10.1007/978-3-030-04221-9_3
その他リンク： https://dblp.uni-trier.de/db/conf/iconip/iconip2018-5.html#ItoSHTT18
-
Reservoir Computing with Untrained Convolutional Neural Networks for Image Recognition 査読あり
Zhiqiang Tong, Gouhei Tanaka
2018 24TH INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION (ICPR) 1289 - 1294 2018年
担当区分：最終著者, 責任著者 記述言語：英語 掲載種別：研究論文（国際会議プロシーディングス） 出版者・発行元：IEEE
Reservoir computing has attracted much attention for its easy training process as well as its ability to deal with temporal data. A reservoir computing system consists of a reservoir part represented as a sparsely connected recurrent neural network and a readout part represented as a simple regression model. In machine learning tasks, the reservoir part is fixed and only the readout part is trained. Although reservoir computing has been mainly applied to time series prediction and recognition, it can be applied to image recognition as well by considering an image data as a sequence of pixel values. However, to achieve a high performance in image recognition with raw image data, a large-scale reservoir including a large number of neurons is required. This is a bottleneck in terms of computer memory and computational cost. To overcome this bottleneck, we propose a new method which combines reservoir computing with untrained convolutional neural networks. We use an untrained convolutional neural network to transform raw image data into a set of smaller feature maps in a preprocessing step of the reservoir computing. We demonstrate that our method achieves a high classification accuracy in an image recognition task with a much smaller number of trainable parameters compared with a previous study.
DOI： 10.1109/ICPR.2018.8545471
その他リンク： https://dblp.uni-trier.de/db/conf/icpr/icpr2018.html#TongT18
-
Reservoir Computing With Spin Waves Excited in a Garnet Film. 査読あり
Ryosho Nakane, Gouhei Tanaka, Akira Hirose
IEEE Access 6 4462 - 4469 2018年
記述言語：英語 掲載種別：研究論文（学術雑誌） 出版者・発行元：Institute of Electrical and Electronics Engineers Inc.
We propose a reservoir computing device utilizing spin waves that propagate in a garnet film equipped with multiple input/output electrodes. In recent years, reservoir computing has been expected to realize energy-efficient and/or high-speed machine learning. Our proposed device enhances such significant merits in a hardware approach. It utilizes the nonlinear interference of history-dependent asymmetrically propagating spin waves excited by the magneto-electric effect. First, we investigate a feasible device structure with practical physical parameters in micromagnetic numerical analysis, and show the detailed characteristics of the forward volume magnetostatic spin waves. Then, we demonstrate high generalization ability in the estimation of input-signal parameters performed by the spin-wave-based reservoir computing. We find that the hysteresis characteristics of the spin waves propagating asymmetrically with respect to excitation points, as well as the nonlinear interference, works advantageously to realize high diversity in the time-sequential signals in high-dimensional information space, which has the highest significance for effective learning in reservoir computing. The spin wave device is highly promising for next-generation machine-learning electronics.
-
Proposal of Carrier-Wave Reservoir Computing. 査読あり
Akira Hirose, Gouhei Tanaka, Seiji Takeda, Toshiyuki Yamane, Hidetoshi Numata, Naoki Kanazawa, Jean Benoit Héroux, Daiju Nakano, Ryosho Nakane
Neural Information Processing - 25th International Conference, ICONIP 2018, Siem Reap, Cambodia, December 13-16, 2018, Proceedings, Part I 616 - 624 2018年
掲載種別：研究論文（国際会議プロシーディングス） 出版者・発行元：Springer
DOI： 10.1007/978-3-030-04167-0_56
その他リンク： https://dblp.uni-trier.de/db/conf/iconip/iconip2018-1.html#HiroseTTYNKHNN18
-
Robustness of coupled oscillator networks with heterogeneous natural frequencies 査読あり 国際誌
Tianyu Yuan, Gouhei Tanaka
CHAOS 27 ( 12 ) 123105 - 123105 2017年12月
担当区分：最終著者, 責任著者 記述言語：英語 掲載種別：研究論文（学術雑誌） 出版者・発行元：AMER INST PHYSICS
Robustness of coupled oscillator networks against local degradation of oscillators has been intensively studied in this decade. The oscillation behavior on the whole network is typically reduced with an increase in the fraction of degraded (inactive) oscillators. The critical fraction of inactive oscillators, at which a transition from an oscillatory to a quiescent state occurs, has been used as a measure for the network robustness. The larger (smaller) this measure is, the more robust (fragile) the oscillatory behavior on the network is. Most previous studies have used oscillators with identical natural frequencies, for which the oscillators are necessarily synchronized and thereby the analysis is simple. In contrast, we focus on the effect of heterogeneity in the natural frequencies on the network robustness. First, we analytically derive the robustness measure for the coupled oscillator models with heterogeneous natural frequencies under some conditions. Then, we show that increasing the heterogeneity in natural frequencies makes the network fragile. Moreover, we discuss the optimal parameter condition to maximize the network robustness.
DOI： 10.1063/1.4991742
-
Interplay between epidemic spread and information propagation on metapopulation networks 査読あり 国際誌
Bing Wang, Yuexing Han, Gouhei Tanaka
JOURNAL OF THEORETICAL BIOLOGY 420 18 - 25 2017年05月
担当区分：最終著者 記述言語：英語 掲載種別：研究論文（学術雑誌） 出版者・発行元：ACADEMIC PRESS LTD- ELSEVIER SCIENCE LTD
The spread of an infectious disease has been widely found to evolve with the propagation of information. Many seminal works have demonstrated the impact of information propagation on the epidemic spreading, assuming that individuals are static and no mobility is involved. Inspired by the recent observation of diverse mobility patterns, we incorporate the information propagation into a metapopulation model based on the mobility patterns and contagion process, which significantly alters the epidemic threshold. In more details, we find that both the information efficiency and the mobility patterns have essential impacts on the epidemic spread. We obtain different scenarios leading to the mitigation of the outbreak by appropriately integrating the mobility patterns and the information efficiency as well. The inclusion of the impacts of the information propagation into the epidemiological model is expected to provide an support to public health implications for the suppression of epidemics.
-
Smoothing effect for spatially distributed renewable resources and its impact on power grid robustness 査読あり 国際誌
Motoki Nagata, Yoshito Hirata, Naoya Fujiwara, Gouhei Tanaka, Hideyuki Suzuki, Kazuyuki Aihara
27 ( 3 ) 033104 - 033104 2017年03月
記述言語：英語 掲載種別：研究論文（学術雑誌）
In this paper, we show that spatial correlation of renewable energy outputs<br />
greatly influences the robustness of power grids. First, we propose a new index<br />
for the spatial correlation among renewable energy outputs. We find that the<br />
spatial correlation of renewable energy outputs in a short time-scale is as<br />
weak as that caused by independent random variables and that in a long<br />
time-scale is as strong as that under perfect synchronization. Then, by<br />
employing the topology of the power grid in eastern Japan, we analyze the<br />
robustness of the power grid with spatial correlation of renewable energy<br />
outputs. The analysis is performed by using a realistic differential-algebraic<br />
equations model and the result shows that the spatial correlation of the energy<br />
resources strongly degrades the robustness of the power grid. Our result<br />
suggests that the spatial correlation of the renewable energy outputs should be<br />
taken into account when estimating the stability of power grids.DOI： 10.1063/1.4977510
