Neural Networks Between Integer and Rational Weights

0470700 - ÚI 2018 RIV US eng C - Konferenční příspěvek (zahraniční konf.)
Šíma, Jiří
Neural Networks Between Integer and Rational Weights.
Proceedings of the 2017 International Joint Conference on Neural Networks. Piscataway: IEEE Operations Center, 2017, s. 154-161. ISBN 978-1-5090-6182-2. ISSN 2161-4407.
[IJCNN 2017. International Joint Conference on Neural Networks /30./. Anchorage (US), 14.05.2017-19.05.2017]
Grant CEP: GA ČR GBP202/12/G061
Institucionální podpora: RVO:67985807
Klíčová slova: computational power * recurrent neural networks * rational weights * cut language * beta-expansion
Obor OECD: Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

The analysis of the computational power of neural networks with the weight parameters between integer and rational numbers is refined. We study an intermediate model of binary-state neural networks with integer weights, corresponding to finite automata, which is extended with an extra analog unit with rational weights, as already two additional analog units allow for Turing universality. We characterize the languages that are accepted by this model in terms of so-called cut languages which are combined in a certain way by usual string operations. We employ this characterization for proving that the languages accepted by neural networks with an analog unit are contextsensitive and we present an explicit example of such non-contextfree languages. In addition, we formulate a sufficient condition when these networks accept only regular languages in terms of quasi-periodicity of parameters derived from their weights.
Trvalý link: http://hdl.handle.net/11104/0268293