WEKO3
インデックスリンク
アイテム
Unlearning of Mixed States in the Hopfield Model : Extensive Loading Case
http://hdl.handle.net/10935/5298
http://hdl.handle.net/10935/52981c74510b-cc10-4afb-8b73-dab2a3128879
| 名前 / ファイル | ライセンス | アクション |
|---|---|---|
aa00704814v87_054004authorsver.pdf
|
|
| アイテムタイプ | default_紀要論文 / Departmental Bulletin Paper(1) | |||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 公開日 | 2019-07-24 | |||||||||||||||||
| タイトル | ||||||||||||||||||
| タイトル | Unlearning of Mixed States in the Hopfield Model : Extensive Loading Case | |||||||||||||||||
| 言語 | ja | |||||||||||||||||
| 言語 | ||||||||||||||||||
| 言語 | jpn | |||||||||||||||||
| キーワード | ||||||||||||||||||
| 言語 | en | |||||||||||||||||
| 主題Scheme | Other | |||||||||||||||||
| 主題 | the Hopfield Model | |||||||||||||||||
| キーワード | ||||||||||||||||||
| 言語 | en | |||||||||||||||||
| 主題Scheme | Other | |||||||||||||||||
| 主題 | Unlearning of Mixed States | |||||||||||||||||
| 資源タイプ | ||||||||||||||||||
| 資源タイプ識別子 | http://purl.org/coar/resource_type/c_6501 | |||||||||||||||||
| 資源タイプ | departmental bulletin paper | |||||||||||||||||
| 著者 |
林,華央
× 林,華央
× 橋本,知奈実
× 木本,智幸× 上江洌,達也 |
|||||||||||||||||
| 抄録 | ||||||||||||||||||
| 内容記述タイプ | Abstract | |||||||||||||||||
| 内容記述 | We study the unlearning of mixed states in the Hopfield model for the extensive loading case. Firstly, we focus on case I, where several embedded patterns are correlated with each other, whereas the rest are uncorrelated. Secondly, we study case II, where patterns are divided into clusters in such a way that patterns in any cluster are correlated but those in two different clusters are not correlated. By using the replica method, we derive the saddle point equations for order parameters under the ansatz of replica symmetry. The same equations are also derived by self-consistent signal-to-noise analysis in case I. In both cases I and II, we find that when the correlation between patterns is large, the network loses its ability to retrieve the embedded patterns and, depending on the parameters, a confused memory, which is a mixed state and/or spin glass state, emerges. By unlearning the mixed state, the network acquires the ability to retrieve the embedded patterns again in some parameter regions. We find that to delete the mixed state and to retrieve the embedded patterns, the coefficient of unlearning should be chosen appropriately. We perform Markov chain Monte Carlo simulations and find that the simulation and theoretical results agree reasonably well, except for the spin glass solution in a parameter region due to the replica symmetry breaking. Furthermore, we find that the existence of many correlated clusters reduces the stabilities of both embedded patterns and mixed states. | |||||||||||||||||
| 言語 | en | |||||||||||||||||
| 内容記述 | ||||||||||||||||||
| 内容記述タイプ | Other | |||||||||||||||||
| 内容記述 | (C) [2018] The Physical Society of Japan (J. Phys. Soc. Jpn. Vol.87, 054004 / 16 pages.) | |||||||||||||||||
| 言語 | ja | |||||||||||||||||
| 書誌情報 |
en : Journal of the Physical Society of Japan 巻 87, p. 54004, 発行日 2018-04-24 |
|||||||||||||||||
| 出版者 | ||||||||||||||||||
| 出版者 | The Physical Society of Japan | |||||||||||||||||
| 言語 | en | |||||||||||||||||
| ISSN | ||||||||||||||||||
| 収録物識別子タイプ | PISSN | |||||||||||||||||
| 収録物識別子 | 1347-4073 | |||||||||||||||||
| DOI | ||||||||||||||||||
| 関連タイプ | isVersionOf | |||||||||||||||||
| 識別子タイプ | DOI | |||||||||||||||||
| 関連識別子 | https://doi.org/10.7566/JPSJ.87.054004 | |||||||||||||||||
| 権利 | ||||||||||||||||||
| 権利情報 | (C) [2018] The Physical Society of Japan (J. Phys. Soc. Jpn. Vol.87, 054004 / 16 pages.) | |||||||||||||||||
| 言語 | en | |||||||||||||||||
| 著者版フラグ | ||||||||||||||||||
| 出版タイプ | VoR | |||||||||||||||||
| 出版タイプResource | http://purl.org/coar/version/c_970fb48d4fbd8a85 | |||||||||||||||||
