Farid Qamar
Data Science || Machine Learning || Remote Sensing || Astrophysics || Public Policy
X
X
+
X
Feed-forward neural network
input layer
hidden layer
output layer
Feed-Forward Neural Network
Recurrent Neural Network
Mon
Tues
Wed
Thurs
Fri
Recurrent Neural Network
Mon
Tues
Wed
Thurs
Fri
O1
Recurrent Neural Network
Mon
Tues
Wed
Thurs
Fri
O1
O2
Recurrent Neural Network
Mon
Tues
Wed
Thurs
Fri
O1
O2
O3
Recurrent Neural Network
Mon
Tues
Wed
Thurs
Fri
O1
O2
O3
O4
Recurrent Neural Network
Mon
Tues
Wed
Thurs
Fri
O1
O2
O3
O4
O5
Recurrent Neural Network
Mon
Tues
Wed
Thurs
Fri
O1
O2
O3
O4
O5
output
Recurrent Neural Network
Recurrent Neural Network
short-term memory!
Final RNN hidden state
Vanishing Gradient Problem!
Final RNN hidden state
Forward Propagation
Forward Propagation
Forward Propagation
Forward Propagation
Forward Propagation
Forward Propagation
Forward Propagation
Forward Propagation
Error Estimation
Forward Propagation
Back Propagation
Error Estimation
Forward Propagation
Back Propagation
Error Estimation
Forward Propagation
Back Propagation
Error Estimation
Forward Propagation
Back Propagation
Error Estimation
Repeat!
Back Propagation
∇
Vanishing Gradient Problem!
∇
Back Propagation
Vanishing Gradient Problem!
∇
Back Propagation
Vanishing Gradient Problem!
∇
Back Propagation
Vanishing Gradient Problem!
∇
Back Propagation
Vanishing Gradient Problem!
∇
Back Propagation
Vanishing Gradient Problem!
∇
Back Propagation
Vanishing Gradient Problem!
∇
Back Propagation
Vanishing Gradient Problem!
∇
Back Propagation
Vanishing Gradient Problem!
∇
Back Propagation
Vanishing Gradient Problem!
∇
Back Propagation
Vanishing Gradient Problem!
1
2Â
3
forget gate
forget irrelevant information
input gate
update gate
add/update new information
pass updated information
LSTM Cell
X
X
+
forget gate
input gate
update gate
sigmoid
tanh
X
pointwise multiplication
pointwise addition
vector concatenation
X
X
+
forget gate
input gate
update gate
sigmoid
tanh
X
pointwise multiplication
pointwise addition
vector concatenation
ht-1
xt
ct-1
X
X
+
forget gate
input gate
update gate
sigmoid
tanh
X
pointwise multiplication
pointwise addition
vector concatenation
ht-1
xt
ct-1
X
X
+
forget gate
input gate
update gate
sigmoid
tanh
X
pointwise multiplication
pointwise addition
vector concatenation
ft
ht-1
xt
ct-1
X
X
+
forget gate
input gate
update gate
sigmoid
tanh
X
pointwise multiplication
pointwise addition
vector concatenation
ht-1
xt
ft
ct-1
X
X
+
forget gate
input gate
update gate
sigmoid
tanh
X
pointwise multiplication
pointwise addition
vector concatenation
ht-1
xt
it
ft
ct-1
X
X
+
forget gate
input gate
update gate
sigmoid
tanh
X
pointwise multiplication
pointwise addition
vector concatenation
ht-1
xt
ft
ct-1
X
X
+
forget gate
input gate
update gate
sigmoid
tanh
X
pointwise multiplication
pointwise addition
vector concatenation
ht-1
xt
ft
ct-1
X
X
+
forget gate
input gate
update gate
sigmoid
tanh
X
pointwise multiplication
pointwise addition
vector concatenation
ht-1
xt
X
X
+
forget gate
input gate
update gate
sigmoid
tanh
X
pointwise multiplication
pointwise addition
vector concatenation
ht-1
xt
X
X
+
forget gate
input gate
update gate
sigmoid
tanh
X
pointwise multiplication
pointwise addition
vector concatenation
ht-1
xt
ct
X
X
+
forget gate
input gate
update gate
sigmoid
tanh
X
pointwise multiplication
pointwise addition
vector concatenation
ht-1
xt
ct
X
X
+
forget gate
input gate
update gate
sigmoid
tanh
X
pointwise multiplication
pointwise addition
vector concatenation
ct
ot
X
X
+
forget gate
input gate
update gate
sigmoid
tanh
X
pointwise multiplication
pointwise addition
vector concatenation
ct
ot
X
X
+
forget gate
input gate
update gate
sigmoid
tanh
X
pointwise multiplication
pointwise addition
vector concatenation
ct
ht
X
X
+
forget gate
input gate
update gate
sigmoid
tanh
X
pointwise multiplication
pointwise addition
vector concatenation
ct
ht
By Farid Qamar