MetaCartSign in to MyCiteSeer

Include Citations | Advanced Search | Help

Include Citations | Advanced Search | Help

  Generalization by structural properties from sparse nested symbolic data

Download:
Download as a PDF
by Mikael Bodén
http://www.hh.se/staff/mibo/publications/sparse2.pdf
Add To MetaCart

Abstract:

Abstract. A set of simulations demonstrate that recurrent networks can exhibit generalization by abstraction from extremely sparse but structurally homogenous symbolic data. By cascading two recurrent networks – feeding the second network with discretized hidden states of the first – it is also possible to generalize according to complex structure. By automatic discretization the cascaded architecture assists in scaling up sequential learning tasks and offers explanations to the apparent systematicity and generativity of language use. 1

Citations

1097 Finding structure in time – Elman - 1988
301 Supervised and unsupervised discretization of continuous features – Dougherty, Kohavi, et al. - 1995
194 The induction of dynamical recognizers – Pollack - 1981
51 A recurrent neural network that learns to count – RODRIGUEZ, WILES, et al. - 1999
50 LSTM recurrent networks learn simple context-free and context-sensitive languages – Gers, A, et al. - 2001
37 Systematicity in connectionist language learning. Mind and Language – Hadley - 1994
25 Context-free and context-sensitive dynamics in recurrent neural networks – Boden, Wiles - 2000
23 Extracting regularities in space and time through a cascade of prediction networks: The case of a mobile robot navigating in a structured environment. Connection Science – Nolfi, Tani - 1999
17 Discovering predictable classifications – Schmidhuber, Prelinger - 1993
4 Language acquisition in the absence of explicit negative evidence: can simple recurrent networks obviate the need for domain-specific learning devices – Marcus - 1999
View or Download | Add to My Collection | Correct Errors