Artificial Neural Networks Generalization and Simplification via Pruning
Scientific Journal 2014 of Riga Technical University
2014
Andrejs Bondarenko,
Arkādijs Borisovs
Artificial neural networks (ANN) are well known for their classification abilities although, but choosing hyper parameters such as neuron layers count and sizes can be quite tedious task. Pruning approaches assume that sufficiently large ANN is already trained and can be simplified with acceptable classification accuracy loss. Current paper presents nodes pruning algorithm and gives experimental results for pruned networks accuracy rates versus their non-pruned counterparts.
Keywords
Artificial Neural Networks, prunning, overfitting, generalization
DOI
10.13140/2.1.4408.5763
Bondarenko, A., Borisovs, A. Artificial Neural Networks Generalization and Simplification via Pruning. Scientific Journal 2014 of Riga Technical University, 2014, Vol. 1, No. 1, pp.9990-9996. Available from: doi:10.13140/2.1.4408.5763
Publication language
English (en)