Training a Single Sigmoidal Neuron Is Hard

DOI10.1162/089976602760408035zbMath1060.68099WikidataQ52112387 ScholiaQ52112387MaRDI QIDQ4409384

Jiří Šíma

Publication date: 14 October 2003

Published in: Neural Computation (Search for Journal in Brave)

Full work available at URL: https://doi.org/10.1162/089976602760408035

zbMATH Keywords

feedforward neural networks; constructive learning; hardness results

Mathematics Subject Classification ID

68T05: Learning and adaptive systems in artificial intelligence

Related Items

Loading Deep Networks Is Hard: The Pyramidal Case, On the Nonlearnability of a Single Spiking Neuron, Long-Range Out-of-Sample Properties of Autoregressive Neural Networks

Cites Work

• Computational limitations on training sigmoid neural networks
• On the complexity of loading shallow neural networks
• On the complexity of polyhedral separability
• Feedforward nets for interpolation and classification
• The densest hemisphere problem
• The hardness of approximate optima in lattices, codes, and systems of linear equations
• Robust trainability of single neurons
• On the geometric separability of Boolean functions
• Learnability and the Vapnik-Chervonenkis dimension
• Computational limitations on learning from examples
• The computational intractability of training sigmoidal neural networks
• Learning representations by back-propagating errors