Averaged Perceptron Algorithm

From GM-RKB
Jump to navigation Jump to search

An Averaged Perceptron Algorithm is a Perceptron Algorithm that is based on the averaged parameters method.



References

2002

Inputs: A training set of tagged sentences, $\left(w^i_{[1:n_i ]} , t^i_{[1:n_i]}\right)$ for $i = 1 \cdots n$. A parameter $T$ specifying number of iterations over the training set. A “local representation” $\Phi$ which is a function that maps history/tag pairs to d-dimensional feature vectors. The global representation $\Phi$ is defined through $\phi$ as in Eq. 1.
Initialization: Set parameter vector $ \overline{\alpha}= 0$.
Algorithm:

For $t = 1 \cdots T, i = 1 \cdot n$

Figure 1: The training algorithm for tagging.