术语或简写备忘 (按字母顺序)
- A/a
activation function 激活函数
artificial neutron: perceptron(以前), sigmoid neutron(如今的模型)
- B/b
backpropagation 反向传播
beta version 测试版本
bias 偏差 :for example, a measure of how easy it is to get the perceptron to output a 11
bitwise 按位
- C/c
carry bit 进位
CNN convolutional neural network 卷积神经网络
convolve 卷积
cost function,is also loss function,objective function
- D/d
Drill down 下钻;沿着维的层次向下,查看更详细的数据,比如从表维度到列维度
Drill up 上钻; 沿着维德层次向上聚集汇总,比如从日销售额到月销售额
钻取可以改变维的层次,变换分析的粒度。
- E/e
elementary logical functions AND, OR, NAND: 基本逻辑函数 与,或,与非(Not AND)
- F/f
feedforward neural networks 前馈神经网络(the output from one layout is an input of the next layer)
- G/g
- H/h
Hadamard product, that is elementwise product(按元素积) ,如
- I/i
incremental learning 渐进式学习,also called online or on-line learning
inner product space 内积空间
greyscale 灰度(图像术语)
- J/j
- K/k
- L/l
- M/m
MSE,mean squared error. is also quadratic cost function
- N/n
NAND gate 与非门:是数字电路的一种基本逻辑电路;(输入全1时,输出为0;只要有一个输入为0,输出就为1;计算例子如下图)
neutron 神经元
neuron firing 神经冲动
NLP natural language processing
notational 符号标注
- O/o
over-fitting 过拟合
- P/p
partial derivatives 偏微分
percetron 感知机
pixels 像素点
pooling 池化
POS part of speech 词性
- Q/q
- R/r
RNN,recurrent neural network (feedback loops)
- S/s
stochastic gradient descend 随机梯度下降
- T/t
translation invariant 平移不变性
- U/u
UDAF user defined aggregate function
UDTF user defined Table-Generating function
- V/v
- W/w
- X/x
- Y/y
- Z/z