Differences

This shows you the differences between two versions of the page.

--- nlp:decoding [2021/03/30 04:05] – created jmflanig
+++ nlp:decoding [2024/08/25 23:53] (current) – [Related Pages] jmflanig
@@ Line 1: / Line 1: @@
 ====== Decoding ======
+In [[structured prediction]] and NLP, //**decoding**// refers to predicting the output given the input.  For example, the input could be a sentence, and the output could be a sequence of part-of-speech tags.  In this case, decoding is predicting the POS tags given the input sentence.  The decoding algorithm is the algorithm used to make the predictions (such as dynamic programming, [[beam search]], [[integer linear programming]], etc).  Decoding is also known as //**inference**// (see note below).
+==== Note on Terminology ====
+In NLP and graphical models, inference means making predictions (predicting the output $Y$ given the input $X$), and is another term for decoding.  In statistics however, inference means learning (i.e. predicting the parameters of the distribution).
 ===== Methods =====
-  * MAP inference in [[ml:probabilistic graphical models|graphical models]]
+  * MAP inference in [[ml:graphical models]]
   * Dynamic Programming
   * [[Weighted Logic Programming]]
@@ Line 8: / Line 12: @@
 ===== Related Pages =====
+  * [[Beam Search]]
   * [[Structured Prediction]]
+  * [[ml:Graphical Models]]
   * [[Integer Linear Programming]]
+  * [[seq2seq#decoding_strategies|Sequence-to-Sequence Models - Decoding Strategies]]
   * [[Weighted Logic Programming]]