期刊名称：Lecture Notes in Engineering and Computer Science
印刷版ISSN：2078-0958
电子版ISSN：2078-0966
出版年度：2018
卷号：2233&2234
页码：374-379
出版社：Newswood and International Association of Engineers
摘要：Many studies have been done with the security of
cloud computing. Though data encryption is a typical approach,
high computing complexity for encryption and decryption of
data is needed. Therefore, safe system for distributed processing
with secure data attracts attention, and a lot of studies have
been done with them. SMC (Secure Multiparty Computation)
is one of these methods. So far, most of works for ML (Machine
Learning) with SMC are ones with supervised and unsupervised
learning such as BP (Back Propagation) and K-means methods.
Then, in the case where learning data does not exist explicitly
like reinforcement learning (RL), how should it be done? We
already proposed some algorithms of Q-learning and PS (Profit
Sharing) learning for SMC in previous papers. However, they
were methods using digital models. On the other hand, solutions
for analog models are desired in the real world. In this paper, we
propose SMC algorithms for Q-learning in the analog model
and show their effectiveness. The idea is that in the digital
model, only one behavior is selected at each time, whereas in
the analog model it is decided as a combination of a plural of
weighted actions.
关键词：cloud computing; secure multiparty computation;;
Q-learning; analog model.
