diff --git a/ml/neural_network/neural_network.md b/ml/neural_network/neural_network.md index ee1214b..d3d9d7f 100644 --- a/ml/neural_network/neural_network.md +++ b/ml/neural_network/neural_network.md @@ -75,7 +75,7 @@ $$ 神经网络是由多个层次的神经元组成的,每个层次含有数量可以是不相同的神经元,一个简单的神经网络结构如下图所示: -![nn_model](images/nn_mode.jpg) +![nn_model](images/nn_model.jpg) 可以看到,任何一个神经网络,都包含一个**输入层**,一个**输出层**,以及若干数量不确定的**隐含层**(`hidden layer`)。输入层一般用$x$或者$a^{(1)}$来表示,其中$x_j$或者$a_j^{(1)}$表示第$j$个特征的值;第一个隐含层用$a^{(2)}$来表示,$a_j^{(2)}$表示第一个隐含层的第$j$个神经元的值,后面的隐含层以此类推;用$L$表示神经网络含有的层次数量,这样输出层就可以用$y$或者$a^{(L)}$来表示了。 @@ -126,7 +126,7 @@ $$ &a^{(1)} = x\\\\ &z^{(2)} = \Theta^{(1)}a^{(1)}, a^{(2)} = sigmoid(z^{(2)})\\\\ &z^{(3)} = \Theta^{(2)}a^{(2)}, a^{(3)} = sigmoid(z^{(3)})\\\\ -&\cdots +&\cdots\\\\ &z^{(L)} = \Theta^{(L - 1)}a^{(L - 1)}, h_\Theta(x) = a^{(L)} = sigmoid(z^{L}) \end{aligned} $$ @@ -147,7 +147,7 @@ $$ 如逻辑回归一样,神经网络也可以被推广到解决多分类问题,两者的方法如出一辙。对于逻辑回归而言,为了解决$n(n \ge 3)$分类问题,只需要训练$n$个二类分类问题的分类器即可;而对于神经网络模型,则只需要将输出层的神经元增加到$n$个,由于前面已经指出,神经网络的最后两个层次本质上就是逻辑回归模型,因此该方法本质上也是训练了$n$个二类分类器,与逻辑回归一致。多分类的神经网络结构如下所示: -![nn_multiclass](images/nn_multicalss.jpg) +![nn_multiclass](images/nn_multiclass.jpg) 但是两者在实现的细节上也有一些差异。对于$n$类分类问题,逻辑回归的输出的取值往往是$\{1, 2, \cdots, n\}$,而神经网络则要沿用向量表示法,例如当$n = 3$时,神经网络的输出应该表示为 diff --git a/words.md b/words.md index 0904f74..f02e0b3 100644 --- a/words.md +++ b/words.md @@ -2194,7 +2194,7 @@ Some Words - Each store is owned by an individual who pays a fee for franchise. - He lost the franchise after failing to meet the specified standards. -##19th, January +## 19th, January + wade > (v)to walk through water or other liquid with some effort, because it is deep enough to come quite high up your legs, or thick. @@ -2219,6 +2219,40 @@ Some Words - It was a clear day with little moisture in the air. - Plants absorb carbon dioxide from the air and moisture from the soil. +## 21st, February + ++ ostensible +> (adj)appearing or claiming to be one thing when it is really something else. + + - Their ostensible goal was to clean up government corruption, but their real aim was to unseat the government. + ++ supersede +> (v)to relace something, especially something older or more old-fashioned. + + - Most of the old roads have been superseded by the great interstate highways. + - Hand tools are relics of the past that have now been superseded by the machine. + ++ disentangle +> (v)to separate things that have become joined or confused. + + - It's difficult to disentangle hard fact from myth, or truth from lies. + - I tried to disentangle the wires under my desk. + +## 23rd, February + ++ commensurate +> (adj)in a correct and suitable amount compared to something else. + + - a salary that is commensurate with skills and experience + - The agency's workload has increased without any commensurate increase in staff. + +## 29th, February + ++ efficacy +> (n)the ability, especially of a medicine or a method of achieving something, to produce the intended result, effectiveness. + + - They recently ran a series of tests to meausre the efficacy of the drug. + + cosmetically + rationale + ameliorate