۱- Introduction

۲- Machine Learning and its Privacy Issues

۳- Tools for Solving the Issues

۴- Secure Cloud-Intelligent Network Framework

۵- Implementation

۶- concLusIon

References

Abstract

Intelligent networks are regarded as existing networks incorporating some intelligent mechanisms such as cognitive and cooperative approaches to improve network performance. Security is highly essential in intelligent networks but has received less attention so far. In this article, we propose a framework that enables a secure intelligent network with the assistance of cloud-assisted privacy-preserving machine learning. In the framework, the cloud server can first generate a model using outsourced machine learning algorithms and then process testing data from the network with the generated model in real time, which reflects to the network and makes it more intelligent. At the same time, the proposal guarantees the security and privacy of both the training data and the testing data in the sense that the proposed framework takes advantage of differential privacy to perform privacy-preserving data analysis and homomorphic encryption to conduct valid operations over encrypted data. The performance evaluations of the core primitives employed in the framework including differential privacy and homomorphic encryption algorithms demonstrate the practicability of our proposal.

Introduction

Having the services loaded into switches in the traditional plain old telephone system complicates the introduction and management of sophisticated services. The growing demand for advanced user-oriented services and the desire to manage the network more cost effectively drive the evolution of a new networking architecture, known as intelligent networking (IN) [1]. IN is essentially an architectural concept for the provision, creation, and management of services that separates the service logic from the underlying physical switching system. The origin of IN can be traced back to 1986 when the basic concept was introduced in the IN/1 definition proposed by Regional Bell Operating Companies. In 1989, the European Telecommunications Standards Institute (ETSI) and the International Telecommunication Union (ITU) began to define the target IN architecture in accordance with the structured development process, aiming to promote the standardization of an international IN. They defined a particular capability set in each phase of evolution. A capability set mainly focuses on two aspects, namely service requirement and network requirement, including service creation, management, interaction, processing, network management, and interworking. The Intelligent Network Conceptual Model (INCM) acts a pivotal part in the process of the target IN architecture, which serves as a complete framework for the design of capability sets. INCM is structured into four layers, and the close interrelation with each other depicting the engineering process of IN is portrayed in Fig. 1. The top layer is the service plane (SP), where users and service providers can describe services without considering their implementation, which is a service-oriented view. The second layer is the global functional plane (GFP), consisting of basic call processing, two interaction points known as point of initiation (POI) and point of return (POR), and a set of service-independent building blocks (SIBs). Each SIB is a unit of functionality, and a chain of SIBs constitutes the service logic described in the SP. The distributed functional plane (DFP) is the third layer, enabling network designers to describe the functional architecture in a distributed view with a range of functional entities (FEs). Any given FE is composed of various functional entity actions (FEAs), and each FEA is performed by a series of elementary functions (EFs). Moreover, a sequence of FEAs and the information flows through them realize the SIB in the second layer. The bottom layer is the physical plane (PP), where multi-equipment vendors model the physical architecture with physical entities (PEs). Each FE from DFP is mapped to one or more PEs, driven by the upper-level service logic. Traditional networking approaches associated with manual, reactive, and centrally administered operations are usually time-consuming and errorprone. However, the next generation network will be large-scale, complex, and heterogeneous. Thus, the traditional networking approaches are unsuitable for the next generation network. Facing the dilemma of data explosion but knowledge shortage, network operators try to optimize the network with the assistance of advanced data analytics such as machine learning (ML) and artificial intelligence (AI), which has attracted much attention from academia and industry so far.