An ELM-Based Privacy Preserving Protocol for Implementing Aware Agents
Hashimoto Masato, Qiangfu Zhao · 2017
Aware agents (A-agents) are systems that can be aware of user intention, preference, situation, etc., and can provide proper solutions. To support user's daily lives, we try to implement A-agents in portable/wearable computing devices (P/WCDs). However, a P/WCD usually does not have enough computing resource and battery to implement high performance A-agents. Cloud computing is a technology to augment the computing power of P/WCDs. But there are two well- known problems, namely, information leakage and privacy invasion. To solve these problems, extreme learning machine (ELM) based privacy preserving protocol has been proposed by us for developing cloud-based A-agents on P/WCDs. The basic idea of the protocol is division of the ELM neural network. The cloud server holds weights of hidden neurons, and the P/WCD holds weights of output neurons. The A-agents are implemented by combining the cloud server and the P/WCD. In this paper, we try to improve the safety of the protocol by adding redundancy. Although adding redundancy can make it more difficult for an unauthorized party to analyze the data and the A-agent model, the computational time cost will be not increased. The required memory will be larger, but this is not a big problem with the current memory technology.