论文部分内容阅读
本文提出了一种采用随机加权联接的新型神经网络模型,该模型具有清晰可分析的神经元激活函数,内部数据表示为随机二进制序列形式,硬件实现十分简洁高效。本文在三个不同层次对前向型网络的学习算法和系统仿真进行了深入的讨论,其中最底层的学习对应于硬件实时在线学习,另外,本文还对系统的泛化性能作了全面的分析,给出了VC维和学习样本量的明确结果。通过对三个用于标准测试的Monk问题和数字手写体识别问题的检测,获得了很好的实验结果。
In this paper, a new neural network model based on stochastic weighted connection is proposed. The model has a clearly analyzed neuron activation function. The internal data is represented as a random binary sequence, and the hardware is very simple and efficient. In this paper, the learning algorithm and system simulation of forward-oriented network are discussed in three different levels. The bottom layer of learning corresponds to the real-time on-line hardware learning. In addition, this paper also analyzes the generalized performance of the system , Gives a clear result of VC dimension and learning sample size. Through the detection of three Monk problems and digital handwriting recognition problems for standard testing, good experimental results have been obtained.