Abstract
The saturation of particular neuron and a whole neural network is one of the reasons for problems with training effectiveness. The paper shows neural network saturation analysis, proposes a method for detection of saturated neurons and its reduction to achieve better training performance. The proposed approach has been confirmed by several experiments.
Author supplied keywords
Cite
CITATION STYLE
Kolbusz, J., Rozycki, P., Lysenko, O., & Wilamowski, B. M. (2018). Neural networks saturation reduction. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 10841 LNAI, pp. 108–117). Springer Verlag. https://doi.org/10.1007/978-3-319-91253-0_11
Register to see more suggestions
Mendeley helps you to discover research relevant for your work.
