I'm currently analysing the paper Fast Lexically Constrained Decoding with Dynamic Beam Allocation for Neural Machine Translation (Post, Vilar 2018): https://arxiv.org/abs/1804.06609
I have understanding problems how the data is processed. For example: the paper is writing about beams, banks and hypothesises and I have no idea what these terms mean.
How would you describe these terms and are there any tutorial sources you would recommend for understanding the dynamic beam allocation?
Beam search is just an optimization technique which selects the top $k$ most probable "solutions" out of a large set of candidate "solutions". You can see it in algorithm 1 in the paper: the function kbest just pick the top $k$ items.
by grouping together hypotheses that have met the same number of constraints into banks
Apparently they use the word "bank" to mean a set of hypothesises grouped together, there's nothing much to it.
How would you describe these terms and are there any tutorial sources you would recommend for understanding the dynamic beam allocation?
In general don't expect tutorials for concepts in research papers, it's very rare that state of the art research becomes mainstream enough for that :)
Usually you have to work your way though the references when you don't understand something.
