caa a22 4500 477034357 CHVBK 20180405111301.0 cr unu---uuuuu 170330e19960401xx s 000 0 eng 10.1007/BF00417786 doi (NATIONALLICENCE)springer-10.1007/BF00417786 Martens Jean-Pierre Electronics and Information Systems Department, University Gent, St.-Pietersnieuwstraat 41, B-9000, Gent, Belgium aut A stochastically motivated random initialization of pattern classifying MLPs [Elektronische Daten] [Jean-Pierre Martens] In this contribution, a new stochastically motivated random weight initialization scheme for pattern classifying Multi-Layer Perceptrons (MLPs) is presented. Its first aim is to ensure that all training examples and all nodes have an equal opportunity to contribute to the improvement of the network during the Error Back-Propagation (EBP) training. In addition, it pursues input scale invariance: if the network inputs were substituted by rescaled inputs, the initialization procedure should provide an equally well performing network. Finally, the new algorithm can initialize MLPs comprising both concentric (e.g., Gaussian) and squashing (e.g., sigmoidal) nodes. Experiments demonstrate that networks initialized using the proposed method train better than networks initialized using a standard random initialization scheme. Kluwer Academic Publishers, 1996 error back-propagation nationallicence network initialization nationallicence pattern recognition nationallicence Neural Processing Letters Kluwer Academic Publishers; http://www.wkap.nl 3/1(1996-04-01), 23-29 1370-4621 3:1<23 1996 3 11063 https://doi.org/10.1007/BF00417786 text/html Onlinezugriff via DOI 1 research-article jats NATIONALLICENCE 856 40 https://doi.org/10.1007/BF00417786 text/html Onlinezugriff via DOI NATIONALLICENCE 100 1- Martens Jean-Pierre Electronics and Information Systems Department, University Gent, St.-Pietersnieuwstraat 41, B-9000, Gent, Belgium aut NATIONALLICENCE 773 0- Neural Processing Letters Kluwer Academic Publishers; http://www.wkap.nl 3/1(1996-04-01), 23-29 1370-4621 3:1<23 1996 3 11063 Metadata rights reserved Springer special CC-BY-NC licence nationallicence BK010053 XK010053 XK010000 NATIONALLICENCE NATIONALLICENCE NL-springer