Inducing high performance neural networks based on an improved decision boundary making algorithm

In recent years, portable computing devices (PCDs) such as smart phones are becoming more and more popular, and many users are using applications on their PCDs. To customize applications for each user, we suggest to use awareness agents (A-agents) that can help users. However, A-agents usually become large. To reduce the size of A-agents, we have proposed decision boundary learning (DBL) based on particle swarm optimization (PSO) algorithm. Through experiments, we can get a compact and high performance A-agent. However, the training time becomes very long. Because, the calculation cost of PSO algorithm is very high. To reduce the calculation cost, we propose a simple method called decision boundary making (DBM) algorithm in this paper. The basic idea of this algorithm is to generate new training data around support vectors (S Vs) of an S VM. Then, an NN is obtained from these new training data. And, for generating data effectively, we set a condition for adding data. Experimental results show that the proposed DBM outperforms DBL, and its learning time is shorter.