神经网络简单概念

Each neural network (NN) is composed of a number of components (layers) joined together by specific connections (synapses). Several neural network architectures can be created (feed forward NN -前馈神经网络, recurrent NN -复发性神经网络, etc) depending on how these components are linked together.
This section deals with feed forward neural networks (FFNN) for simplicity's sake, but it is possible to build whatever neural network architecture is required with Joone.
为简单起见本节讨论前馈神经网络（FFNN），不过依托Joone构建任意要求的神经网络都是可能的。

A FFNN is composed of a number of consecutive layers with each one connected to the next by a synapse. In a FFNN recurrent connections from a layer to a previous one are not permitted. Consider the following figure:
FFNN是由许多彼此依靠神经突触相互连接的的连续的层组成的。在一个FFNN中一个层与其前驱层是不允许循环连接的。如下图所示：

 
：一对一

This is a sample FFNN with two layers fully connected with synapses. Each layer is composed of a certain number of neurons, each of which have the same characteristics (transfer function, learning rate, etc).
这是一个简单的拥有突触全连接的两层的FFNN。每层有确定数量的神经元组成，每个神经元都有相同的特性（传递功能，学习速率等）。

A neural net built with Joone can be composed of any number of layers belonging to different typologies (linear, sigmoid, etc).
以Joone构建的神经网络可以是任意数量的属于不同类型的层组成的（线性，S型等）。

Each layer processes its input signal by applying a transfer function and sending the resulting pattern to the synapses that connect it to the next layer. So a neural network can process an input pattern and transferring that pattern from an input layer to an output layer.
每层通过应用传递功能处理它的输入信号并发送结果模式给连接着它和下一层的神经突触。因此，一个神经网络可以处理来自输入层的输入模式并传输该模式到输出层。

This is the basic concept upon which the entire engine is based.
这是整个引擎基于的基本概念。