Neural networks with values in multidimensional domains have been intensively studied over the last few years. This paper introduces octonion-valued neural networks, for which the inputs, outputs, weights and biases are all octonions. They represent a generalization of the complex- and quaternion-valued neural networks, that do not fall into the category of Clifford-valued neural networks, because, unlike Clifford algebras, the octonion algebra is not associative. The full deduction of the gradient descent algorithm for training octonion-valued feedforward neural networks is presented. Testing of the proposed network is done using two synthetic function approximation problems and a time series prediction application.
CITATION STYLE
Popa, C. A. (2016). Octonion-valued neural networks. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 9886 LNCS, pp. 435–443). Springer Verlag. https://doi.org/10.1007/978-3-319-44778-0_51
Mendeley helps you to discover research relevant for your work.