講題：Distributed Consensus Reduced Support Vector Machine

時間：05/14(二) 13:20

地點：南大校區推廣教育大樓9407教室

摘要：Nowadays, machine learning performs astonishingly in many

      different fields. The more data we have, our machine learning

      methods will show better results. However, in some cases, the

      data owners may not want to or not allow to share the data they

      have. On the other hand, we may encounter extremely large data

      sets that even cannot be stored in a single machine. In order

      to deal with these two problems, we propose the distributed

      consensus reduced support vector machine (DCRSVM) for binary

      classification. Image that we have a set of local working units

      and one center master. The DCSVM allows the local working units

      share the local models without sharing their own data.

      Iteratively, by sharing and updating the local models, the center

      master will generate a consensus final model. The performance of

      the consensus model is approximately as good as the model trained

      by using all local working units’ data together. Similarly,

      training an extremely large dataset, we can divide the dataset

      into many partitions and dispatch the partitions to many

      computation units. Thus, our proposed method can satisfy the

      requirement of no data sharing.