Abstract

Sequence prediction is a popular machine learning task, consisting of making predictions about future symbols based on a previously observed sequence. This makes it ideal for NLP tasks such as machine translation, where the ordering of the symbols (words) affects the meaning of the text.

In this talk we will look at how sequence-to-sequence models can be emulated using the Encoder-Decoder architecture. This approach will involve two recurrent neural networks, one to encode the input sequence (Encoder) and the other to decode the encoded input sequence into the target sequence (Decoder). We will then consider how this model can be applied to machine translation, and some other common use cases.