Adaptive Context Vector Generation For Each Decoding Step

* 🏆 2014: [Adam: A Method for Stochastic Optimization (Adam)](https://arxiv.org/abs/1412.6980)

Unique: Generates a fresh context vector at each decoding step by attending to source annotations, rather than using a single fixed context vector, enabling the decoder to dynamically select relevant source information based on what it has already generated

vs others: Adaptive context vectors enable better translation of long sentences and complex reorderings vs fixed-context encoder-decoder, because the model can attend to different source regions for different target positions