A controllable IPv6 address generation based on conditional variational autoencoder

Abstract

Address generation is one of the important methods to improve the efficiency of network assert detection, cyberspace mapping and so on. Most existing address generations consider all IPv6 addresses in IPv6 address space, which wastes resource and cannot meet requirements in practice. In this paper, a new address generation based on conditional variational autoencoder is proposed. Specifically, BiLSTM is embedded in conditional variational autoencoder to extract sequential information among addresses. With the given address region, seed addresses are encoded by the improved conditional variational autoencoder, which can generate addresses in a given region and make network assert detection do in a specific region in practice. Experimental results show that the generation ratio and the concentration of the proposed model improve 5% to 15% and 60% to 80% respectively.