thecenter

Activities and Seminars

Paul Boersma. A single neural network for supervised and unsupervised learning
 
Date: Dec 17, 2012

What: A single neural network for supervised and unsupervised learning

Where: BCBL auditorium

Who: Paul Boersma, University of Amsterdam (in collaboration with Titia Benders and Klaas Seinhorst)

When: 10.30am


A single artificial neural network structure (an auditory continuum connected bidirectionally with a competitive phonological layer) with a single learning algorithm (inoutstar) is capable of learning new phonological categories from distributional auditory input without supervision, while with supervision from the lexicon it is capable of producing realistic auditory dispersion effects. I will show a variety of computer simulations illustrating the concepts and the effects.