Recurrent Neural Networks
Today Recurrent Neural Network Cell Recurrent Neural Networks (unenrolled) LSTMs, Bi-LSTMs, Stacked Bi-LSTMs
Recurrent Neural Network Cell 𝑅𝑁𝑁 ℎ 0 ℎ 1 𝑥 1
Recurrent Neural Network Cell ℎ 1 =tanh( 𝑊 ℎℎ ℎ 0 + 𝑊 ℎ𝑥 𝑥 1 ) 𝑅𝑁𝑁 ℎ 0 ℎ 1 𝑥 1
Recurrent Neural Network Cell 𝑦 1 ℎ 1 𝑅𝑁𝑁 ℎ 0 ℎ 1 ℎ 1 =tanh( 𝑊 ℎℎ ℎ 0 + 𝑊 ℎ𝑥 𝑥 1 ) 𝑥 1 𝑦 1 =softmax( 𝑊 ℎ𝑦 ℎ 1 )
Recurrent Neural Network Cell 𝑦 1 ℎ 1 𝑅𝑁𝑁 ℎ 0 ℎ 1 𝑥 1
Recurrent Neural Network Cell 𝑦 1 =[0.1, 0.05, 0.05, 0.1, 0.7] ℎ 1 =[1 2 0 3 0 0 1 ] 𝑅𝑁𝑁 ℎ 0 =[0 0 0 0 0 0 0 ] 𝑥 1 = [0 0 1 0 0] a b c d e c
Generating Samples from the Recurrent Neural Network Cell
LSTM Cell (Long Short-Term Memory) ℎ 0 ℎ 1 𝐿𝑆𝑇𝑀 𝑐 0 𝑐 1 𝑥 1
Recurrent Neural Network Cell 𝑦 1 ℎ 1 𝑅𝑁𝑁 ℎ 0 ℎ 1 𝑥 1
Recurrent Neural Network Cell ℎ 1 𝑅𝑁𝑁 ℎ 0 ℎ 1 𝑥 1
(Unrolled) Recurrent Neural Network <<space>> 𝑦 1 ℎ 1 ℎ 2 ℎ 3 ℎ 0 𝑅𝑁𝑁 ℎ 1 𝑅𝑁𝑁 ℎ 2 𝑅𝑁𝑁 ℎ 3 𝑥 1 𝑥 2 𝑥 3 c a t
(Unrolled) Recurrent Neural Network cat likes eating 𝑦 1 ℎ 1 ℎ 2 ℎ 3 ℎ 0 𝑅𝑁𝑁 ℎ 1 𝑅𝑁𝑁 ℎ 2 𝑅𝑁𝑁 ℎ 3 𝑥 1 𝑥 1 𝑥 1 the cat likes
(Unrolled) Recurrent Neural Network positive / negative sentiment rating 𝑦 1 ℎ 3 ℎ 0 𝑅𝑁𝑁 ℎ 1 𝑅𝑁𝑁 ℎ 2 𝑅𝑁𝑁 ℎ 3 𝑥 1 𝑥 1 𝑥 1 the cat likes
(Unrolled) Recurrent Neural Network 𝑥 1 𝑅𝑁𝑁 ℎ 0 ℎ 1 ℎ 2 ℎ 3 c a t <<space>> 𝑦 1
Bidirectional Recurrent Neural Network gato quiere comer 𝑦 1 ℎ 1 ℎ 2 ℎ 3 ℎ 0 𝐵𝑅𝑁𝑁 ℎ 1 B𝑅𝑁𝑁 ℎ 2 𝐵𝑅𝑁𝑁 ℎ 3 𝑥 1 𝑥 1 𝑥 1 the cat wants
Stacked Recurrent Neural Network 𝑦 1 𝑦 1 𝑦 1 ℎ 1 ℎ 2 ℎ 3 ℎ 0 𝑅𝑁𝑁 ℎ 1 𝑅𝑁𝑁 ℎ 2 𝑅𝑁𝑁 ℎ 3 ℎ 1 ℎ 2 ℎ 3 ℎ 0 𝑅𝑁𝑁 ℎ 1 𝑅𝑁𝑁 ℎ 2 𝑅𝑁𝑁 ℎ 3 𝑥 1 𝑥 1 𝑥 1 c a t
Bidirectional Stacked Recurrent Neural Network 𝑦 1 𝑦 1 ℎ 1 ℎ 2 ℎ 3 ℎ 0 𝑅𝑁𝑁 ℎ 1 𝑅𝑁𝑁 ℎ 2 𝑅𝑁𝑁 ℎ 3 ℎ 1 ℎ 2 ℎ 3 ℎ 0 𝑅𝑁𝑁 ℎ 1 𝑅𝑁𝑁 ℎ 2 𝑅𝑁𝑁 ℎ 3 𝑥 1 𝑥 1 𝑦 1 𝑥 1 c a t
Questions?