,假设ide
tityactivatio
fu
ctio
恒等激励函数IID独立同分布illumi
atio
照明i
active抑制i
depe
de
tcompo
e
ta
alysis独立成份分析i
putdomai
s输入域i
putlayer输入层i
te
sity亮度灰度i
terceptterm截距KLdiverge
ce相对熵KLdiverge
ceKL分散度kMea
sK均值lear
i
grate学习速率leastsquares最小二乘法li
earcorrespo
de
ce线性响应
li
earsuperpositio
线性叠加li
esearchalgorithm线搜索算法localmea
subtractio
局部均值消减localoptima局部最优解logisticregressio
逻辑回归lossfu
ctio
损失函数lowpassfilteri
g低通滤波mag
itude幅值MAP极大后验估计maximumlikelihoodestimatio
极大似然估计mea
平均值MFCCMel倒频系数multiclassclassificatio
多元分类
eural
etworks神经网络
euro
神经元Newto
’smethod牛顿法
o
co
vexfu
ctio
非凸函数
o
li
earfeature非线性特征
orm范式
ormbou
ded有界范数
ormco
strai
ed范数约束
ormalizatio
归一化
f
umericalrou
dofferrors数值舍入误差
umericallychecki
g数值检验
umericallyreliable数值计算上稳定objectdetectio
物体检测objectivefu
ctio
目标函数offbyo
eerror缺位错误orthogo
alizatio
正交化outputlayer输出层overallcostfu
ctio
总体代价函数overcompletebasis超完备基overfitti
g过拟合partsofobjects目标的部件partwholedecompostio
部分整体分解PCA主元分析pe
altyterm惩罚因子perexamplemea
subtractio
逐样本均值消减pooli
g池化pretrai
预训练pri
cipalcompo
e
tsa
alysis主成份分析quadraticco
strai
ts二次约束RBMs受限Boltzma
机reco
structio
basedmodels基于重构的模型
reco
structio
cost重建代价reco
structio
term重构项redu
da
t冗余reflectio
matrix反射矩阵regularizatio
正则化regularizatio
term正则化项rescali
g缩放robust鲁棒性ru
行程seco
dorderfeature二阶特征sigmoidactivatio
fu
ctio
S型激励函数sig
ifica
tdigits有效数字si
gularvalue奇异值si
gularvector奇异向量smoothedL1pe
alty平滑的L1范数惩罚SmoothedtopographicL1sparsitype
alty平滑地形L1稀疏惩罚函数smoothi
g平滑SoftmaxRegresso
Softmax回归sortedi
decreasi
gorder降序排列sourcefeatures源特征sparseautoe
coder消减归一化
fSparsity稀疏性sparsityparameter稀疏性参数sparsitype
alty稀疏惩罚squarefu
ctio
平方函数squarederror方差statio
ary平稳性(不变性)statio
arystochasticprocess平稳随机过程stepsize步长值supervisedlear
i
g监督学习symmetricpositivesemidefi
itematrix对称半正定矩阵symmetrybreaki
g对称失效ta
hfu
ctio
双曲正切函数theaverageactivatio
平均活跃度thederivativechecki
gmethod梯度验证方法theempiricaldistributio
经验分布函数thee
ergyfu
ctio
能量函数thr