Encode-Decoder
Models
How do we translate the following phrase to a different language?
All models are wrong, some are useful.
Todos los modelos son incorrectos, algunos son útiles.
AASAHFTGCPWVURDYU
How would a computer translate for one sequence to another?
What are the important things that a computer must know?
Encoder-Decoder models is designed to take in a sequence, conduct mathematical functions, and transform it to a new sequence. Commonly known as seq2seq problems.
The Encoder component is designed to handle the input sequence and model it with numerical values.
The numerical values are stored in both long-term and short-term memories, LSTM cells.
The Decoder component takes the inputs from the long-term and short-term memory of the encoder component and inputs it into a new LSTM cell
The LSTM cell will then predict the first value (word) of a sequence.
The word perdicted will then be used for a new LSTM cell to predict the next word.
The process continues until an <EOS> token (end of sequence) is produced.
m408.inqs.info/lectures/12a