LSTM解决RNN梯度消失与梯度爆炸问题

RNN(Recurrent Neural Network)由于其递归的网络结构(如图1所示),对于处理序列建模任务具有独特的优势,因此在许多领域有着广泛的应用。如自然语言处理、语音识别等。

1.RNN的BPTT

LSTM解决RNN梯度消失与梯度爆炸问题_第1张图片
根据RNN的网络结构可写出其基本方程:
S t = δ ( W S t − 1 + U X t )         ( 1 ) O t = δ ( V S t )                         ( 2 ) S_{t} = \delta(WS_{t-1} + UX_{t}) \ \ \ \ \ \ \ (1) \\ O_{t} = \delta(VS_{t}) \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ (2) St=δ(WSt1+UXt)       (1)Ot=δ(VSt)                       (2)
假设交叉熵为其损失函数loss:
L = − ∑ t = 1 n O t l o g O t ^                       ( 3 ) L=-\sum_{t=1}^{n}O_{t}log\hat{O_{t}} \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ (3) L=t=1nOtlogOt^                     (3)
然后分别对W、U、V求偏导
先求V的偏导,因其偏导较为简单
∂ L ∂ V = ∂ L ∂ O t ⋅ ∂ O t ∂ V                  ( 4 ) \frac{\partial L}{\partial V}=\frac{\partial L}{\partial O_{t}}\cdot \frac{\partial O_{t}}{\partial V} \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ (4) VL=OtLVOt                (4)
再对W和U求偏导
由公式(1)可知,当前时刻的状态不仅与当前的输入有关,而且还与与前一时刻的状态有关。
对W和U运用链式求导
∂ L ∂ W = ∂ L ∂ O t ⋅ ∂ O t ∂ S t ⋅ ∂ S t ∂ S t − 1 ⋅ ∂ S t − 1 ∂ S t − 2 ⋅ . . . ⋅ ⋅ ∂ S 1 ∂ S 0 ⋅ ∂ S 0 ∂ W = ∂ L ∂ O t ⋅ ∂ O t ∂ S t ⋅ ∏ k = 1 t ∂ S k ∂ S k − 1 ⋅ ∂ S k − 1 ∂ W       ( 5 ) \begin{aligned} \frac{\partial L}{\partial W}&=\frac{\partial L}{\partial O_{t}}\cdot \frac{\partial O_{t}}{\partial S_{t}}\cdot \frac{\partial S_{t}}{\partial S_{t-1}}\cdot \frac{\partial S_{t-1}}{\partial S_{t-2}}\cdot...\cdot \cdot \frac{\partial S_{1}}{\partial S_{0}}\cdot \frac{\partial S_{0}}{\partial W}\\ &=\frac{\partial L}{\partial O_{t}}\cdot \frac{\partial O_{t}}{\partial S_{t}}\cdot \prod_{k=1}^{t} \frac{\partial S_{k}}{\partial S_{k-1}}\cdot \frac{\partial S_{k-1}}{\partial W}\ \ \ \ \ (5) \end{aligned} WL=OtLStOtSt1StSt2St1...S0S1WS0=OtLStOtk=1tSk1SkWSk1     (5)
同理可得
∂ L ∂ U = ∂ L ∂ O t ⋅ ∂ O t ∂ S t ⋅ ∂ S t ∂ S t − 1 ⋅ ∂ S t − 1 ∂ S t − 2 ⋅ . . . ⋅ ⋅ ∂ S 1 ∂ S 0 ⋅ ∂ S 0 ∂ U = ∂ L ∂ O t ⋅ ∂ O t ∂ S t ⋅ ∏ k = 1 t ∂ S k ∂ S k − 1 ⋅ ∂ S k − 1 ∂ U       ( 6 ) \begin{aligned} \frac{\partial L}{\partial U}&=\frac{\partial L}{\partial O_{t}}\cdot \frac{\partial O_{t}}{\partial S_{t}}\cdot \frac{\partial S_{t}}{\partial S_{t-1}}\cdot \frac{\partial S_{t-1}}{\partial S_{t-2}}\cdot...\cdot \cdot \frac{\partial S_{1}}{\partial S_{0}}\cdot \frac{\partial S_{0}}{\partial U}\\ &=\frac{\partial L}{\partial O_{t}}\cdot \frac{\partial O_{t}}{\partial S_{t}}\cdot \prod_{k=1}^{t} \frac{\partial S_{k}}{\partial S_{k-1}}\cdot \frac{\partial S_{k-1}}{\partial U}\ \ \ \ \ (6) \end{aligned} UL=OtLStOtSt1StSt2St1...S0S1US0=OtLStOtk=1tSk1SkUSk1     (6)

2.RNN梯度消失与梯度爆炸

由公式(1)可知
∂ S t ∂ S t − 1 = W ⋅ σ ′       ( 7 ) \frac{\partial S_{t}}{\partial S_{t-1}}=W\cdot {\sigma }'\ \ \ \ \ (7) St1St=Wσ     (7)
sigmod函数
LSTM解决RNN梯度消失与梯度爆炸问题_第2张图片
当公式(7)的乘积小于1时,公式(5)和公式(6)就会趋近于0,也即梯度消失;
当公式(7)的乘积大于1时,公式(5)和公式(6)就会趋近于无穷大,也即梯度爆炸;

3.LSTM解决RNN梯度问题

LSTM解决RNN梯度消失与梯度爆炸问题_第3张图片
PS:图片来源于http://colah.github.io/posts/2015-08-Understanding-LSTMs/ 下面公式中的标号参考该链接中图片标号。
i t = σ ( W i [ h t − 1 ; x t ] + b i )         ( 8 ) f t = σ ( W f [ h t − 1 ; x t ] + b f )        ( 9 ) C ~ t = t a n h ( W c [ h t − 1 ; x t ] + b c )    ( 10 ) C t = i t ∗ C ~ t + f t ∗ C t − 1          ( 11 ) o t = σ ( W o [ h t − 1 ; x t ] + b o )        ( 12 ) h t = o t ∗ t a n h ( C t )                   ( 13 ) \begin{aligned} i_{t}&=\sigma (W_{i}[h_{t-1}; x_{t}]+b_{i}) \ \ \ \ \ \ \ (8) \\ f_{t}&=\sigma (W_{f}[h_{t-1}; x_{t}]+b_{f}) \ \ \ \ \ \ (9) \\ \tilde{C}_{t}&=tanh (W_{c}[h_{t-1}; x_{t}]+b_{c}) \ \ (10) \\ C_{t}&=i_{t}*\tilde{C}_{t}+f_{t}*C_{t-1} \ \ \ \ \ \ \ \ (11) \\ o_{t}&=\sigma (W_{o}[h_{t-1}; x_{t}]+b_{o}) \ \ \ \ \ \ (12) \\ h_{t}&=o_{t}*tanh(C_{t}) \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ (13) \\ \end{aligned} itftC~tCtotht=σ(Wi[ht1;xt]+bi)       (8)=σ(Wf[ht1;xt]+bf)      (9)=tanh(Wc[ht1;xt]+bc)  (10)=itC~t+ftCt1        (11)=σ(Wo[ht1;xt]+bo)      (12)=ottanh(Ct)                 (13)
类比RNN中偏导的连乘部分,LSTM中连乘部分为
∂ C t ∂ C t − 1 = f t = σ           ( 14 ) \frac{\partial C_{t}}{\partial C_{t-1}}=f_{t}=\sigma \ \ \ \ \ \ \ \ \ (14) Ct1Ct=ft=σ         (14)
对比公式(7)和公式(14),LSTM的连乘部分变成了σ,在实际参数更新过程中,通过控制其值接近于1,则经过多次连乘(训练)后,梯度也不会消失;而σ的值不会大于1,故不会出现梯度爆炸。

你可能感兴趣的:(NLP,DeepLearning,基础算法,LSTM,RNN,梯度消失,梯度爆炸)