Visual comparison of performance for different activation functions in MLP networks
作者： F. Piekniewski , L. Rybicki
DOI: 10.1109/IJCNN.2004.1381133
关键词:
摘要: Multi layer perceptron networks have been successful in many applications, yet there are unsolved problems the theory. Commonly, sigmoidal activation functions used, giving good results. The backpropagation algorithm might work with any other function on one condition though - it has to a differential. We investigate some possible and compare results they give sample data sets.
sci-hub.se 参考文章(6)
Filip Piȩkniewski, Leszek Rybicki, Visualizing and Analyzing Multidimensional Output from MLP Networks via Barycentric Projections international conference on artificial intelligence and soft computing. pp. 247- 252 ,(2004) , 10.1007/978-3-540-24844-6_33
Pierre Peretto, An introduction to the modeling of neural networks ,(1992)
Simon Haykin, Neural Networks: A Comprehensive Foundation ,(1998)
W. Duch, Uncertainty of data, fuzzy membership functions, and multilayer perceptrons IEEE Transactions on Neural Networks. ,vol. 16, pp. 10- 23 ,(2005) , 10.1109/TNN.2004.836200
W. Duch, Coloring black boxes: visualization of neural network decisions international joint conference on neural network. ,vol. 3, pp. 1735- 1740 ,(2003) , 10.1109/IJCNN.2003.1223669
Pierre Peretto, An Introduction to the Modeling of Neutral Networks Cambridge University Press. ,(1992) , 10.1017/CBO9780511622793