activation 激活值 

activation function 激活函數(shù) 

additive noise 加性噪聲 

autoencoder 自編碼器 

Autoencoders 自編碼算法 

average firing rate 平均激活率 

average sum-of-squares error 均方差 

backpropagation 后向傳播 

basis 基 

basis feature vectors 特征基向量 

batch gradient ascent 批量梯度上升法 

Bayesian regularization method 貝葉斯規(guī)則化方法 

Bernoulli random variable 伯努利隨機(jī)變量 

bias term 偏置項(xiàng) 

binary classfication 二元分類 

class labels 類型標(biāo)記 

concatenation 級聯(lián) 

conjugate gradient 共軛梯度 

contiguous groups 聯(lián)通區(qū)域 

convex optimization software 凸優(yōu)化軟件 

convolution 卷積 

cost function 代價函數(shù) 

covariance matrix 協(xié)方差矩陣 

DC component 直流分量 

decorrelation 去相關(guān) 

degeneracy 退化 

demensionality reduction 降維 

derivative 導(dǎo)函數(shù) 

diagonal 對角線 

diffusion of gradients 梯度的彌散 

eigenvalue 特征值 

eigenvector 特征向量 

error term 殘差 

feature matrix 特征矩陣 

feature standardization 特征標(biāo)準(zhǔn)化 

feedforward architectures 前饋結(jié)構(gòu)算法 

feedforward neural network 前饋神經(jīng)網(wǎng)絡(luò) 

feedforward pass 前饋傳導(dǎo) 

fine-tuned 微調(diào) 

first-order feature 一階特征 

forward pass 前向傳導(dǎo) 

forward propagation 前向傳播 

Gaussian prior 高斯先驗(yàn)概率 

generative model 生成模型 

gradient descent 梯度下降 

Greedy layer-wise training 逐層貪婪訓(xùn)練方法 

grouping matrix 分組矩陣 

Hadamard product 阿達(dá)馬乘積 

Hessian matrix Hessian 矩陣 

hidden layer 隱含層 

hidden units 隱藏神經(jīng)元 

Hierarchical grouping 層次型分組 

higher-order features 更高階特征 

highly non-convex optimization problem 高度非凸的優(yōu)化問題 

histogram 直方圖 

hyperbolic tangent 雙曲正切函數(shù) 

hypothesis 估值，假設(shè) 

identity activation function 恒等激勵函數(shù) 

IID 獨(dú)立同分布 

illumination 照明 

inactive 抑制 

independent component analysis 獨(dú)立成份分析 

input domains 輸入域 

input layer 輸入層 

intensity 亮度/灰度 

intercept term 截距 

KL divergence 相對熵 

KL divergence KL分散度