Deep clustering learns deep feature representations that solve clustering tasks with a deep autoencoder. However, blurred clustering is inevitable due to the lack of overall clustering environment information, which means that no significant differences were observed between clusters. To address this issue, we propose a memory enhanced model for deep clustering with reinforcement strategy, where a Memory Cell is introduced as a storage unit for surrounding imformation. Specifically, the whole process of the model is divided into three parts. Firstly, the original high-dimensional image data is mapped to latent feature space thorugh the pre-training process, and the latent feature representation is obtained and stored in Memory Cell. Secondly, the traditional K-means algorithm is applied to initialize the clustering center on the latent representation. Finally, the reward regression strategy in reinforcement learning based on the Bernoulli distribution is adopted to fine-tune the results. ACC, ARI and NMI as evaluation metrics, the proposed model shows its competitiveness on MNIST and Fashion-MNIST dataset against recent state-of-art models.
{"title":"Deep Clustering with Reinforcement Strategy","authors":"Chenxin Liu","doi":"10.1145/3478301.3478311","DOIUrl":"https://doi.org/10.1145/3478301.3478311","url":null,"abstract":"Deep clustering learns deep feature representations that solve clustering tasks with a deep autoencoder. However, blurred clustering is inevitable due to the lack of overall clustering environment information, which means that no significant differences were observed between clusters. To address this issue, we propose a memory enhanced model for deep clustering with reinforcement strategy, where a Memory Cell is introduced as a storage unit for surrounding imformation. Specifically, the whole process of the model is divided into three parts. Firstly, the original high-dimensional image data is mapped to latent feature space thorugh the pre-training process, and the latent feature representation is obtained and stored in Memory Cell. Secondly, the traditional K-means algorithm is applied to initialize the clustering center on the latent representation. Finally, the reward regression strategy in reinforcement learning based on the Bernoulli distribution is adopted to fine-tune the results. ACC, ARI and NMI as evaluation metrics, the proposed model shows its competitiveness on MNIST and Fashion-MNIST dataset against recent state-of-art models.","PeriodicalId":338866,"journal":{"name":"The 2nd European Symposium on Computer and Communications","volume":"11 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-04-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125694358","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：