Before 2015 when the first attention model was proposed, machine translation was based on the simple encoder-decoder model, a stack of  RNN and LSTM layers. The encoder is used to process the entire sequence of input data into a context vector. This is expected to be a good summary of input data. The final stage of the encoder is the initial stage of the decoder. 

The post Hands-on Guide to Effective Image Captioning Using Attention Mechanism appeared first on Analytics India Magazine.