Magri to appear: How to keep the HG weights non-negative: the truncated Perceptron reweighing rule

Direct link: http://roa.rutgers.edu/content/article/files/1507_giorgio_magri_1.pdf

ROA: 1268
Title: How to keep the HG weights non-negative: the truncated Perceptron reweighing rule
Authors: Giorgio Magri
Comment: to appear in the Journal of Language Modelling
Length: 32
Abstract: The literature on error-driven learning in Harmonic Grammar (HG) has adopted the Perceptron reweighing rule. Yet, this rule is not suited to HG, as it fails at ensuring non-negative weights. A variant is thus considered which truncates the updates at zero, keeping the weights non-negative. Convergence guarantees and error bounds for the original Perceptron are shown to extend to its truncated variant.
Type: Paper/tech report
Keywords: learnability, error-driven learning, HG, Perceptron

 

Facebooktwitterredditpinterestlinkedinmail