Cyrus Shaoul: "NDL: An R package for large-scale, naïve, discriminative learning."

Описание: Cyrus Shaoul speaking at annual meeting of the Society for Computers in Psychology (SCIP), November 2013 in Toronto.

Abstract
We introduce naive discriminative learning (NDL) model,
based on the Rescorla-Wagner equations, and simulated using the
equilibria approximation proposed by Danks (2003). NDL allows the
consequences of accumulated experience in lexical processing to be
explored with realistic data samples. This means learning to use
thousands of cues to discriminate tens of thousands of lexemes, with
training on corpora ranging from tens of millions of words to several
billions of words. However more training data is not necessarily
better. For example, from an information-theoretic perspective,
entropy increases with vocabulary/corpus size, and the model explains
how processing speeds inevitably slow down as exeprience grows, such
that the model fits empirical aging data with no "decline". We also
present results from recent work predicting eye-movements and EEG
waveforms using weights derived from our NDL models. Our
implementation of NDL is a package that is freely available for the
open-source R statistical computing environment.

More about Cyrus at https://www.cyr.us/