One of the reasons I like neural network theory is that it provides real explanations for cognitive phenomena, not black boxes. If someone says 'working memory' in a neural network theory, they need to provide the actual neural network structure or set of algorithms, not just a hand-drawn box on a document. And so we have a case in point here with dreams. As the author notes, " most cognitive theories treat dreams as an epiphenomenon." But as outlined in this article (and reference publication by Erik Hoel (18 page PDF)) dreams may have evolved because they address overfitting in neural networks, and hence prevent us from generalizing on insignificant details.
Today: 5 Total: 95 [Direct link] [Share]
