MISC - 田中 剛平
-
リザバーコンピューティング 招待あり
田中剛平
電子情報通信学会誌 106 ( 6 ) 2023年06月
-
リザバーコンピューティングの概念と最近の動向 (小特集 リザバーコンピューティング)—Concept of Reservoir Computing and Its Recent Trends
田中 剛平
電子情報通信学会誌 = The journal of the Institute of Electronics, Information and Communication Engineers 102 ( 2 ) 108 - 113 2019年02月
-
スピン波リザバーコンピューティング:実用的な計算性能向上の手法
中根了昌, 廣瀬明, 田中剛平
応用物理学会春季学術講演会講演予稿集(CD-ROM) 70th 2023年03月
-
Guest Editorial Special Issue on New Frontiers in Extremely Efficient Reservoir Computing
Gouhei Tanaka, Claudio Gallicchio, Alessio Micheli, Juan Pablo Ortega, Akira Hirose
IEEE Transactions on Neural Networks and Learning Systems 33 ( 6 ) 2571 - 2574 2022年06月
-
磁区を伝播するスピン波を用いたリザバーコンピューティング
中根了昌, 廣瀬明, 田中剛平, 田中剛平, 田中剛平
応用物理学会春季学術講演会講演予稿集(CD-ROM) 69th 2022年
-
スパイキングニューラルネットワークとreward-modulated STDPによるリザバーコンピューティング
鶴海杭之, 田中剛平, 田中剛平
電子情報通信学会技術研究報告(Web) 122 ( 65(NLP2022 1-25) ) 2022年
-
リザバーコンピューティングによる再構成処理を利用した時系列信号の異常検知の試み
加藤准也, 田中剛平, 田中剛平, 中根了昌, 廣瀬明
電子情報通信学会技術研究報告(Web) 121 ( 390(NC2021 46-78) ) 2022年
-
リザバー状態のマハラノビス距離を用いた省メモリ型時系列異常検知
田村浩人, 田中剛平, 田中剛平, 藤原寛太郎
日本神経化学会大会抄録集(Web) 65th 2022年
-
リワードマシンを用いる強化学習手法の計算性能とタスク難易度の関係
渡邊隆二, 田中剛平
電子情報通信学会技術研究報告(Web) 121 ( 444(NLP2021 126-152) ) 2022年
-
教師なしパターン認識のためのスパイキングニューラルネットワークにおけるスパース結合の影響
品川大樹, 藤原寛太郎, 田中剛平, 田中剛平
電子情報通信学会技術研究報告(Web) 121 ( 444(NLP2021 126-152) ) 2022年
-
ニューラルネットワークハードウエアとスピン波リザバーコンピューティング
廣瀬明, 田中剛平, 中根了昌
日本物理学会講演概要集(CD-ROM) 76 ( 1 ) 2021年
-
スピン波リザバーコンピューティングチップにおける読み出し電極配置の規則性が性能に与える影響
市村剛大, 中根了昌, 田中剛平, 廣瀬明
日本神経回路学会全国大会講演論文集 31st (CD-ROM) 2021年
-
超低消費電力情報処理実現のための物理リザバーコンピューティングチップと複素ニューラルネットワーク理論
廣瀬明, 中根了昌, 田中剛平
応用物理学会秋季学術講演会講演予稿集(CD-ROM) 81st 214 - 214 2020年08月
-
知っておきたいキーワード(第136回)リザバーコンピューティング—Keywords you should know : Reservoir Computing
田中 剛平
映像情報メディア学会誌 = The journal of the Institute of Image Information and Television Engineers 74 ( 3 ) 532 - 534 2020年05月
-
メタ個体群モデルを用いた感染症対策の解析
松木 彩星, 田中 剛平
生産研究 72 ( 2 ) 129 - 135 2020年03月
記述言語:英語 出版者・発行元:東京大学生産技術研究所
感染症の数理モデルは,複雑な感染症伝播過程の理解や感染症対策の有効性評価に用いられる.その中で,メタ個体群感染症モデルは,空間的に離れたパッチ(区画)に分布した個体群とパッチ間の個体の移動を考慮するモデルである.このような現実的要素を取り入れるモデルでありながら,ある程度の理論的解析も可能である.本稿では,一部のパッチに介入して(つまり感染症対策を施して)実効的な感染率を引き下げるような局所的対策に注目し,感染症の大規模流行を予防するために必要な最小介入割合(介入閾値)の理論的導出や選択的介入の有効性を検証するための数値シミュレーションについて解説する.
-
リザバーコンピューティングの数理とデバイス実装
田中 剛平
応用物理学会学術講演会講演予稿集 2020.1 49 - 49 2020年02月
-
ネットワークと脳科学 (特集 ネットワークから見る世界 : 数理でつながりを読み解く)
田中 剛平, 合原 一幸
数理科学 58 ( 2 ) 59 - 65 2020年02月
-
スピン波リザバーコンピューテイングにおける有用情報の空間分布
市村剛大, 中根了昌, 中根了昌, 田中剛平, 田中剛平, 廣瀬明, 廣瀬明
電子情報通信学会技術研究報告 118 ( 470(NC2018 44-88)(Web) ) 1 - 8 2020年
-
パターン情報表現およびパターン情報処理を物理的に実現するニューラルネットワークデバイス
廣瀬明, 廣瀬明, 田中剛平, 田中剛平, 武田征士, 山根敏志, 沼田秀俊, 金澤直輝, HEROUX J. B, 中野大樹, 中根了昌, 中根了昌
電子情報通信学会大会講演論文集(CD-ROM) 2019 ROMBUNNO.CS‐2‐7 2019年03月
-
スピン波を用いたリザバーコンピューティングデバイスにおける荷重の空間分布
市村剛大, 中根了昌, 中根了昌, 田中剛平, 田中剛平, 廣瀬明, 廣瀬明, 廣瀬明
電子情報通信学会大会講演論文集(CD-ROM) 2019 ROMBUNNO.CS‐2‐8 2019年03月
-
パターン情報表現およびパターン情報処理を物理的に実現するニューラルネットワークデバイスとその利点
廣瀬明, 田中剛平, 武田征士, 山根敏志, 沼田秀俊, 金澤直輝, HEROUX J. B, 中野大樹, 中根了昌
電子情報通信学会技術研究報告 118 ( 470(NC2018 44-88)(Web) ) 117‐120 (WEB ONLY) 2019年02月
-
スピン波リザバーコンピューティングチップデバイス
中根了昌, 田中剛平, 廣瀬明
応用物理学会春季学術講演会講演予稿集(CD-ROM) 66th ROMBUNNO.12p‐W933‐10 2019年02月
-
「ニューロ的」ハードウエアと物理リザバーコンピューティング
廣瀬明, 武田征士, ヘロー ジャンベノ, 金澤直輝, 沼田英俊, 山根敏志, 中野大樹, 中根了昌, 田中剛平
日本神経回路学会全国大会講演論文集 29th 2019年
-
光波による物理リザバーコンピューテイング
廣瀬明, 武田征士, HEROUX Jean Benoit, 沼田英俊, 金澤直輝, 山根敏行, 中野大樹, 中根了昌, 田中剛平
Optics & Photonics Japan講演予稿集(CD-ROM) 2019 2019年
-
スピン波を用いた機械学習チップデバイス (特集 ニューロデバイスの研究最前線)—On-Chip Machine-Learning Device with Spin Waves
中根 了昌, 田中 剛平, 廣瀬 明
まぐね = Magnetics Japan 14 ( 6 ) 329 - 334 2019年
-
スピン波リザバーコンピューティングチップにおける実空間情報分布
市村剛大, 中根了昌, 田中剛平, 廣瀬明, 中根了昌, 田中剛平, 廣瀬明
日本神経回路学会全国大会講演論文集 29th 2019年
-
招待講演 複雑ネットワークの動的頑強性 (複雑コミュニケーションサイエンス)—Dynamical robustness of complex networks
田中 剛平
電子情報通信学会技術研究報告 = IEICE technical report : 信学技報 118 ( 316 ) 33 - 38 2018年11月
-
複雑ネットワークの動的頑強性
田中剛平
電子情報通信学会技術研究報告 118 ( 316(CCS2018 33-45) ) 33‐38 2018年11月
-
多段階学習Echo state networkによる非線形時系列予測
秋山貴則, 田中剛平, 田中剛平
電子情報通信学会技術研究報告 118 ( 284(IBISML2018 44-104)(Web) ) 293‐299 (WEB ONLY) 2018年10月
-
ハードウェアで実現するリザバーコンピューティング
中野大樹, 沼田英俊, 山根敏志, 武田征士, HEROUX Jean Benoit, 金澤直樹, 田中剛平, 中根了昌, 廣瀬明
電気学会全国大会講演論文集(CD-ROM) 2018 ROMBUNNO.S16‐4 2018年03月
-
大規模停電を想定した電力系統モデルにおけるカスケード故障 (非線形問題)
瀧澤 雄登, 田中 剛平
電子情報通信学会技術研究報告 = IEICE technical report : 信学技報 117 ( 121 ) 63 - 66 2017年07月
-
リザーバ・コンピューティングにおける複素ニューラルネットワーク
廣瀬明, 武田征士, 田中剛平, 中根了昌, 山根敏志, 中野大樹, 中川茂
電子情報通信学会大会講演論文集(CD-ROM) 2017 2017年
-
興奮性振動子結合系における自発的な振動現象
森野 佳生, 田中 剛平, 合原 一幸
日本物理学会講演概要集 72.1 ( 0 ) 3035 - 3035 2017年
記述言語:日本語 出版者・発行元:一般社団法人 日本物理学会
<p>今回の発表では,興奮性を示す位相振動子を結合した系において,結合強度を強化することで起こる自発的振動現象の解析結果について報告する.今回の系では結合強度を上げることで安定平衡点がサドルノード分岐により消失して自発的な振動が生じるが,全ての素子が振動を開始する場合と一部の素子のみが振動を開始する場合がパラメータにより分かれる.この振動開始のメカニズムを分岐解析によって詳細に解析した結果を報告する.</p>
-
外部フィードバック・レーザを用いたリザーバコンピューティング
武田征士, 山根敏志, 中野大樹, 田中剛平, 中根了昌, 廣瀬明, 中川茂
電子情報通信学会大会講演論文集(CD-ROM) 2017 2017年
-
非線形物理ダイナミクスによるリザーバ・コンピューティング
山根敏志, 武田征士, 中野大樹, 田中剛平, 中根了昌, 廣瀬明, 中川茂
電子情報通信学会大会講演論文集(CD-ROM) 2017 2017年
-
ニューロン特性の不均一性を考慮したエコーステートネットワーク
田中剛平, 中根了昌, 山根敏志, 中野大樹, 武田征士, 中川茂, 廣瀬明
電子情報通信学会大会講演論文集(CD-ROM) 2017 2017年
-
細胞システムの動態と分岐理論
田中 剛平
生物物理 56 ( 6 ) 340 - 344 2016年
-
A pruning method based on weight variation information for feedforward neural networks
Zhiqiang Tong, Gouhei Tanaka
IFAC-PapersOnLine 28 ( 18 ) 221 - 226 2015年11月
記述言語:英語
Artificial neural networks are powerful tools for many information processing tasks such as pattern recognition, data mining, optimization, and prediction. It is a significant problem to find optimal structures of artificial neural networks for drawing out their high computational performance. Downsizing of network structure is also an issue to be considered for hardware implementation of largescale neural networks. In this study, we propose a pruning method to find a compact structure of feedforward neural networks with high generalization ability in classification problems. Our method evaluates the significance of neuron nodes using the information on weight variation in the training process and prune the insignificant nodes preferentially unless the classification accuracy is degraded. Numerical experiments with several benchmark datasets show that the proposed method is effective compared with other methods.
-
A hybrid model for hepatitis B virus
Ben-Gong Zhang, Gouhei Tanaka, Kazuyuki Aihara, Masao Honda, Shuichi Kaneko, Luonan Chen
IFAC-PapersOnLine 28 ( 18 ) 37 - 40 2015年11月
記述言語:英語
The hepatitis B virus (HBV) is a kind of pandemic infectious disease. It is becoming a major health problem in Asian and African countries, however, less so in other areas around the world. There is a vast background for studying HBV. However, some problems still require further study. For example, when the patient is infected by HBV, how to arrange the treatment schedule? Using long term continuous treatment or others? Therefore, we wish to readdress the study of HBV. First, we propose a new hybrid model that contains both continuous and discrete variables. Second, we analyze its dynamical behavior. We hope that this study may provide some new insight for HBV disease and some new guidance for the therapy.
-
Dynamical robustness of complex biological networks
Gouhei Tanaka, Kai Morino, Kazuyuki Aihara
Mathematical Approaches to Biological Systems: Networks, Oscillations, and Collective Motions 29 - 53 2015年01月
記述言語:英語 出版者・発行元:Springer Japan
Dynamical behavior of biological systems is maintained by interactions between biological units such as neurons, cells, proteins, and molecules. It is a challenging issue to understand robustness of biological interaction networks from a viewpoint of dynamical systems. In this chapter, we introduce the concept of dynamical robustness in complex networks and demonstrate its application to biological networks. First, we introduce the framework for studying the dynamical robustness through analyses of coupled Stuart-Landau oscillators with various types of network structures. Second, based on the framework, we examine the dynamical robustness of neuronal firing activity in networks of synaptically coupled Morris- Lecar neuron models. Our analyses suggest that a consideration of both network structure and dynamics is crucial in elucidating biological robustness.
-
21aPS-5 ヘテロ性を持つ周期性・興奮性位相振動子結合系の頑強性
森野 佳生, 田中 剛平, 合原 一幸
日本物理学会講演概要集 70.1 ( 0 ) 2853 2015年
-
17pPSA-52 ヘテロ性を持つ周期性・興奮性位相振動子結合系の複雑な相図の構造について
森野 佳生, 田中 剛平, 合原 一幸
日本物理学会講演概要集 70.2 ( 0 ) 2704 2015年
-
Wave-Based Neuromorphic Computing Framework for Brain-Like Energy Efficiency and Integration
Y. Katayama, T. Yamane, D. Nakano, R. Nakane, G. Tanaka
2015 IEEE 15TH INTERNATIONAL CONFERENCE ON NANOTECHNOLOGY (IEEE-NANO) 1210 - 1213 2015年
記述言語:英語 出版者・発行元:IEEE
We present a framework of wave-based neuromorphic computing aiming at brain-like capabilities and efficiencies with nano-scale device integration. We take advantage of the unique nature of elastic wave dynamics in both computations and the IO communications in between as a means to efficiently implement and execute neuromorphic computing. We point out that the group velocity of a wave is one of the essential design parameters.
-
Wave-Based Device Scaling Concept for Brain-Like Energy Efficiency and Integration
Y. Katayama, T. Yamane, D. Nakano, R. Nakane, G. Tanaka
PROCEEDINGS OF THE 2015 IEEE/ACM INTERNATIONAL SYMPOSIUM ON NANOSCALE ARCHITECTURES (NANOARCH 15) 23 - 24 2015年
記述言語:英語 出版者・発行元:IEEE
Aiming at achieving brain-like capabilities and efficiencies with nano-scale integration, this paper proposes a new scaling concept for wave-based neuromorphic devices. Our scaling approach shows that v(g) the group velocity of a signaling wave is one of the essential parameters that can account for the orders-of-magnitude gaps in the efficiency of the human brain and VLSIs for cognitive tasks.
-
疎結合を持つモジュラー型自己連想ニューラルネットワークの特性解析 (情報論的学習理論と機械学習 情報論的学習理論ワークショップ)—Performance analysis of auto-associative neural networks on diluted modular networks
山根 敏志, 田中 剛平, 中野 大樹
電子情報通信学会技術研究報告 = IEICE technical report : 信学技報 114 ( 306 ) 351 - 356 2014年11月
記述言語:日本語 出版者・発行元:一般社団法人電子情報通信学会
モジュラー・ネットワーク構造を持つ自己連想ニューラルネットワークの特性解析について報告する.特に,ハードウェア実装を念頭に置き,配線の複雑さを低減するために各モジュールは疎な結合を有する場合を考察する.学習アルゴリズムは任意のネットワーク・トポロジに適用可能な局所反復型のHebb則を用い,モジュラー構造と自己連想の特性の関係について議論する.
-
Hopfield-type associative memory with sparse modular networks
Gouhei Tanaka, Toshiyuki Yamane, Daiju Nakano, Ryosho Nakane, Yasunao Katayama
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 8834 255 - 262 2014年01月
© Springer International Publishing Switzerland 2014. Modular structures are ubiquitously found in the brain and neural networks. Inspired by the biological networks, we explore Hopfieldtype recurrent neural networks with sparse modular connectivity for associative memory. We first show that an iterative learning algorithm, which determines the connection weights depending on the network topology, yields better performance than the one-shot learning rule. We then examine the topological factors which govern the memory capacity of the sparse modular neural network. Numerical results suggest that the uniformity of the number of connections per neuron is an essential condition for good performance. We discuss a method to design an energy-efficient neural network.
-
On the analysis of the bifurcation sets of equilibrium points in parameter space
G. Chesi, G. Tanaka, Y. Hirata, K. Aihara
2013 European Control Conference, ECC 2013 670 - 675 2013年12月
This paper addresses the problems of characterizing and estimating the bifurcation sets of equilibrium points in multi-parameter space of a class of nonlinear dynamical systems. Specifically, we investigate the sets of parameters that lead to saddle-node bifurcations and Hopf bifurcations at an equilibrium point of interest. First, a characterization of these sets is provided in terms of the zeros of some functions. Second, this characterization is exploited to estimate such sets through convex programming for the case of polynomial dynamical systems. In particular, two conditions are proposed for establishing whether a sublevel set of a given polynomial does not contain parameters that lead to bifurcations. By using these conditions, the largest of such sublevel sets can be estimated by solving an eigenvalue problem. Some numerical examples illustrate the proposed results. © 2013 EUCA.
-
28aPS-55 活性・不活性位相振動子を選択的に配置した複雑ネットワークにおける振動現象(領域11ポスターセッション,領域11(統計力学,物性基礎論,応用数学,力学,流体物理))
森野 佳生, 田中 剛平, 合原 一幸
日本物理学会講演概要集 68 ( 0 ) 284 2013年
-
27pXM-6 大域結合振動子系における揺らぎの非標準的なスケーリング則(27pXM 反応拡散系・振動子系・力学系とその周辺,領域11(統計力学,物性基礎論,応用数学,力学,流体物理))
西川 功, 田中 剛平, 合原 一幸
日本物理学会講演概要集 68.1.2 ( 0 ) 359 2013年
-
28aPS-36 ワクチン接種の不均一性を考慮した風しん流行の数理モデル(領域11ポスターセッション,領域11(統計力学,物性基礎論,応用数学,力学,流体物理))
占部 千由, 田中 剛平, 光藤 哲也, 三崎 貴子, 丸山 絢, 大嶋 孝弘, 岡部 信彦, 合原 一幸
日本物理学会講演概要集 68 ( 0 ) 281 2013年
-
位相モデルを用いた東日本送電網における周波数同期の解析
永田 基樹, 藤原 直哉, 西川 功, 田中 剛平, 鈴木 秀幸, 合原 一幸
生産研究 65 ( 3 ) 295 - 299 2013年
記述言語:日本語 出版者・発行元:東京大学生産技術研究所
東日本大震災以降,電力の安定供給は重要な課題として注目を集めている.<br>安定性の理解のためには数理モデルの解析が必要であるが,詳細なモデルは数理的な解析が容易ではないため,現象の本質を抽出した簡潔なモデルが必要である.本研究では,東日本の送電網の接続関係を用いて,周波数同期のダイナミクスを位相モデルによって記述した.本モデルにおいて,安定供給のためには同期した定常状態の実現が必要であるが,そのための結合強度の閾値を求めた.また,周波数同期に寄与している送電線を特定し,同期への寄与の大きさについて考察した.
-
インフルエンザ流行モデルにおける動的ワクチン配分の最適制御問題 (特集 工学とバイオ)
山本 浩貴, 田中 剛平, 合原 一幸
生産研究 64 ( 3 ) 313 - 318 2012年05月
-
Dynamical modeling of chronic myeloid leukemia progression and the development of mutations
Ben-gong Zhang, Gouhei Tanaka, Luonan Chen, Kazuyuki Aihara
2012 IEEE INTERNATIONAL CONFERENCE ON BIOINFORMATICS AND BIOMEDICINE WORKSHOPS (BIBMW) 130 - 135 2012年
記述言語:英語 出版者・発行元:IEEE
Mathematical models have already been developed to describe the dynamics of the chronic myeloid leukemia (CML). However, most of them mainly focus on the short term dynamics and response to the treatment with drugs such as imatinib or the relapse upon cessation of treatment. In this paper, we aim to study the dynamics of CML which represents the first human cancer for the molecularly targeted therapy again. First, we propose a new mathematical model of CML which based on the known biology know ledges. And then we analyze its qualitative and dynamical properties. Combining the clinical data and theoretical analysis, we demonstrate that our model is biologically plausible and also computationally viable. Second, we demonstrate that this model can characterize the dynamics of the CML infection and the development of mutation, and is also validated by clinical data.
-
1SC-04 生命ネットワークにおける動的ロバスト性の数理的解析(1SC 生物学における数学的手法の最前線,シンポジウム,日本生物物理学会第50回年会(2012年度))
田中 剛平, Morino Kai, Aihara Kazuyuki
生物物理 52 ( 0 ) S3 - S4 2012年
-
21aAE-10 活性・不活性位相振動子の複雑ネットワークにおける振動現象の次数依存性(21aAE 反応拡散系・振動子系,領域11(統計力学,物性基礎論,応用数学,力学,流体物理))
森野 佳生, 田中 剛平, 合原 一幸
日本物理学会講演概要集 67 ( 0 ) 309 2012年
-
インフルエンザ流行モデルにおける動的ワクチン配分の最適制御問題
山本 浩貴, 田中 剛平, 合原 一幸
生産研究 64 ( 3 ) 313 - 318 2012年
記述言語:日本語 出版者・発行元:東京大学生産技術研究所
新型インフルエンザが流行の兆しを見せはじめたときには,感染拡大を抑えるために迅速にワクチンの製造・配分計画をたてて実施することが重要である.インフルエンザワクチンの製造はウィルスの抗原型が特定されてから行われるため,流行初期に十分な量のワクチンを配布することは困難な場合が多く,その時々で使用可能なワクチン数を考慮しながら感染率の異なる宿主集団に対して動的にワクチン配分を行う必要がある.本研究では,年齢構造化インフルエンザ流行モデルを用いて,総死亡者数や総感染者数を最小化するワクチン資源の動的配分問題を考え,最適制御問題として定式化した.最適制御アルゴリズムを用いてこの問題を解くことにより,感染被害を最小化するためには時期に応じてどの年齢層に優先的にワクチンを配分すればよいかが明らかとなる.また,この動的ワクチン配分の最適制御は,新型および季節性インフルエンザの同時流行モデルに対しても応用できることを示す.[本要旨はPDFには含まれない]
-
複雑系数理モデル学とシミュレーション
田中 剛平, 奥 牧人, 合原 一幸
シミュレーション 30 ( 4 ) 236 - 241 2011年12月
-
Analysis of a learning algorithm for a single phasor neuron (Research on complex dynamics and related fields)
田中 剛平
数理解析研究所講究録 1762 ( 1762 ) 1 - 12 2011年09月
-
Complex-sigmoid function for complex-valued neural networks (力学系--理論から応用へ、応用から理論へ--RIMS研究集会報告集)
田中 剛平
数理解析研究所講究録 1742 ( 1742 ) 145 - 157 2011年05月
-
Mathematical modelling of prostate cancer growth and its application to hormone therapy (vol 368, pg 5029, 2010)
Gouhei Tanaka, Yoshito Hirata, S. Larry Goldenberg, Nicholas Bruchovsky, Kazuyuki Aihara
PHILOSOPHICAL TRANSACTIONS OF THE ROYAL SOCIETY A-MATHEMATICAL PHYSICAL AND ENGINEERING SCIENCES 369 ( 1935 ) 495 - 495 2011年01月
-
25pTF-2 層構造を持つ振動子ネットワーク上でのエイジング転移(25pTF 反応拡散系・振幅方程式系,領域11(統計力学,物性基礎論,応用数学,力学,流体物理))
森野 佳生, 田中 剛平, 合原 一幸
日本物理学会講演概要集 66 ( 0 ) 296 2011年
-
23pGU-4 結合振動子ネットワークにおける大域的振動の回復について(23pGU 反応拡散系・振幅方程式系,領域11(統計力学,物性基礎論,応用数学,力学,流体物理))
森野 佳生, 田中 剛平, 合原 一幸
日本物理学会講演概要集 66 ( 0 ) 286 2011年
-
1周期と2周期のリミットサイクル振動子の結合系における個数比の影響に関する数値的解析
岡田 祐典, 田中 剛平, 河野 崇, 合原 一幸
電子情報通信学会技術研究報告. NLP, 非線形問題 109 ( 269 ) 203 - 208 2009年11月
記述言語:日本語 出版者・発行元:一般社団法人電子情報通信学会
結合振動子系は,例えば細胞集団が互いに相互作用して生成する生体リズムの仕組みを調べるための数理モデルとして,広く研究されてきた.本研究では,個々の振動子が不均一であるような結合振動子系におけるネットワークの挙動を考察する.特に,個々の振動子は同じ力学系として与えられるが,システムパラメータはそれぞれ異なる場合を考える.大同ら(2004)は,リミットサイクルに収束するようなパラメータ値をもつactive素子と平衡点に収束するようなパラメータ値をもつinactive素子が混在するときのネットワークダイナミクスを解析し,総振動子数のうちinactive振動子が占める割合が変化したときの相転移(aging transition)を考察した.この研究に動機を得て,本研究では,結合がないときに1周期リミットサイクルに収束する振動子と2周期リミットサイクルに収束する振動子を大域的に拡散結合したネットワークの挙動を数値計算により調べ,拡散結合強度と1周期振動子の占有率が変化したときの相図を示す.
-
Backpropagation Learning Algorithm for Multi layer Phasor Neural Networks
Gouhei Tanaka, Kazuyuki Aihara
NEURAL INFORMATION PROCESSING, PT 1, PROCEEDINGS 5863 484 - 493 2009年
記述言語:英語 出版者・発行元:SPRINGER-VERLAG BERLIN
We present a backpropagation learning algorithm for multi-layer feedforward phasor neural networks using a gradient descent method. The state of a phasor neuron takes a complex-valued state on the unit circle in the complex domain. Namely, the state can be identified only by its phase component because the amplitude component is fixed. Due to the circularity of the phase variable, phasor neural networks are useful to deal with periodic and multivalued variables. Under the assumption that the weight coefficients are complex numbers and the activation function is a continuous and differentiable function of a phase variable, we derive an iterative learning algorithm to minimize the output error. In each step of the algorithm, the weight coefficients are updated in the gradient descent direction of the error function landscape. The proposed algorithm is numerically tested in function approximation task. The numerical results suggest that the proposed method has a better generalization ability compared with the other backpropagation algorithm based on linear correction rule.
-
感染症流行モデルによる感染力を低下させる予防手段の効果に関する試算
田中 剛平, 合原 一幸
生産研究 61 ( 6 ) 1081 - 1084 2009年
記述言語:日本語 出版者・発行元:東京大学生産技術研究所
新型インフルエンザが世界的に流行し, 今後さらに拡大が続くと懸念されている.今冬以降の流行に備えてワクチンの製造が急ピッチで進められているが, その製造量には限界があるため, ワクチン接種以外の予防手段を徹底することも感染拡大を防止するのに重要だと考えられている.本研究では, 患者のマスク着用や外出の自重, 部屋の換気, 非感染者の手洗いやうがいの徹底などの予防手段によって新型インフルエンザの感染力が一定の割合で低下すると仮定し, その効果について感染症流行モデルを用いて試算した.その結果, 例えば1人の感染者が生み出す2次感染者数の平均を1.4人とした場合, 上記の対策により感染力を10%減らすことができれば, その効果はワクチン約1600万人分に相当し, また最終的な罹患率を約25%減らせることが分かった.[本要旨はPDFには含まれない]
-
カオス工学は何を可能にするか (特集 非線形という科学の見方)
田中 剛平
科学 78 ( 11 ) 1242 - 1245 2008年11月
-
非線形多値関数を用いた複素連想記憶モデル
田中 剛平, 合原 一幸
電子情報通信学会技術研究報告. NLP, 非線形問題 107 ( 478 ) 13 - 18 2008年02月
記述言語:日本語 出版者・発行元:一般社団法人電子情報通信学会
複素ニューラルネットワークは,ニューロンの状態を適当な閾値関数によって離散多値化することで,多値要素から成るパターンの連想記憶を行うことができる.従来の多値連想記憶モデルで用いられてきた複素閾値関数は,実質的に多値符号化関数を含む変換である.本稿では,これを非線形多値関数で置き換えた複素閾値関数を提案する.記憶パターンベクトルを一般化射影則によって結合重みに埋め込み,提案した閾値関数をニューロンの状態更新に用いると,多値連想記憶の能力が向上することを示す.また,非線形性を制御するパラメータ,記憶パターン数,ニューロンがとり得る離散値の数などが提案手法の性能に与える影響を数値実験により明らかにする.さらに,提案手法が多階調画像のノイズ除去に応用できることを実例によって示す.
-
MP-533 前立腺癌の間欠的ホルモン療法の数理モデルによる解析(一般演題ポスター,第94回日本泌尿器科学会総会)
島田 尚, 合原 一幸, 出田 亜位子, 津本 国親, 辻 繁樹, 平田 祥人, 鈴木 大慈, 田中 剛平, 武内 功, 北村 唯一
日本泌尿器科学会雑誌 97 ( 2 ) 500 2006年
-
Multistate associative memory with parametrically coupled map networks
Gouhei Tanaka, Kazuyuki Aihara
IEEE Asia-Pacific Conference on Circuits and Systems, Proceedings, APCCAS 2 1017 - 1020 2004年12月
The present paper proposes two types of multistate associative memory models using circle maps coupled through a parameter in the individual map. Each network utilizes a circle map with a specific bifurcation property as its component and realizes self-organizing chaotic dynamics in a memory association. The performance of the proposed networks is compared with that of a conventional multistate neural network in multistate associative memory tests. © 2004 IEEE.
-
2次元写像を用いたバーストセルモデルの解析
田中 剛平, 合原 一幸
電子情報通信学会技術研究報告. CAS, 回路とシステム 103 ( 334 ) 73 - 78 2003年09月
記述言語:日本語 出版者・発行元:一般社団法人電子情報通信学会
生体内の興奮性細胞の膜電位にみられるバースト応答は,細胞間情報伝達において重要な役割を持つと言われている.バースト応答を示す細胞を2つ電気シナプスによって結合したとき,結合係数パラメータに応じてバースト応答の同位相または逆位相同期状態が安定となることが,複数の実験及びモデルシミュレーションによって報告されている.そこで本報告では,単一のバースト細胞として単純な2次元写像モデルを使用し,電気シナプス結合系においてバースト同期が同位相から逆位相へと変化するメカニズムを調べる.結合強度パラメータの符号を変えることによって,不変部分空間の局所的な横断的安定性が切り替わり,結合が正のとき同位相解が,結合が負のとき逆位相解が安定となることを示す.また,単一細胞と結合系において,バースト応答が見られるパラメータ領域を比較する.