自组织增量学习神经网络综述

自组织增量学习神经网络SOINN（self-organizing incremental neural network）是一种基于竞争学习的两层神经网络，用于在没有先验知识的情况下对动态输入数据进行在线聚类和拓扑表示，同时，对噪音数据具有较强的鲁棒性.SOINN的增量性，使得它能够发现数据流中出现的新模式并进行学习，同时不影响之前学习的结果.因此，SOINN能够作为一种通用的学习算法应用于各类非监督学习问题中.对SOINN的模型和算法进行相应的调整，可以使其适用于监督学习、联想记忆、基于模式的推理、流形学习等多种学习场景中.SOINN已经在许多领域得到了应用，包括机器人智能、计算机视觉、专家系统、异常检测等.

Review of Self-Organizing Incremental Neural Network

Self-organizing incremental neural network (SOINN) is a two layered, competitive learning based neural network which is able to represent the topology structure of input data and cluster online non-stationary data without prior knowledge, and also robust to noise. The incremental nature of SOINN enables it to learn novel patterns from data stream without affecting previously learned patterns. In this respect, it is appropriate to expect that SOINN could serve as a general approach to unsupervised learning problems. With some modifications, SOINN could handle other kinds of learning tasks such as supervised learning, associative memory, pattern based reasoning and manifold learning as well. SOINN has been used in many kinds of applications including robotics, computer vision, expert systems, and anomaly detection. This paper presents a survey of its basic ideas, improvements and applications.