Training Neural Networks with Local Error Signals

20 Jan 2019Arild NøklandLars Hiller Eidnes

Supervised training of neural networks for classification is typically performed with a global loss function. The loss function provides a gradient for the output layer, and this gradient is back-propagated to hidden layers to dictate an update direction for the weights... (read more)

PDF Abstract

Evaluation results from the paper


Task Dataset Model Metric name Metric value Global rank Compare
Image Classification CIFAR-10 VGG11B(2x)+LocalLearning+CO Percentage correct 96.4 # 15
Image Classification CIFAR-10 VGG11B(2x)+LocalLearning+CO Percentage error 3.60 # 11
Image Classification CIFAR-100 VGG11B(3x)+LocalLearning Percentage correct 79.9 # 14
Image Classification CIFAR-100 VGG11B(3x)+LocalLearning Percentage error 20.1 # 7
Image Classification Fashion-MNIST VGG8B(2x)+LocalLearning+CO Percentage error 4.14 # 2
Image Classification Kuzushiji-MNIST VGG8B+LocalLearning+CO Percentage error 0.99 # 1
Image Classification MNIST VGG8B+LocalLearning+CO Percentage error 0.26 # 3
Image Classification STL-10 VGG8B+LocalLearning+CO Percentage correct 80.75 # 5
Image Classification SVHN VGG8B+LocalLearning+CO Percentage error 1.65 # 8