search

dyigitpolat / mimarsinan

a pipeline for modeling in-memory architectures for spiking neural networks
MIT License
6 stars 0 forks source link

mixed use of training and validation accuracies #40

Closed dyigitpolat closed 1 year ago

dyigitpolat commented 1 year ago

mixed use of training and validation accuracies causes the adaptation process to work in unpredictable ways especially when training and validation accuracies are different because of "overfitting" or certain "regularization" methods.

this issue is linked to https://github.com/dyigitpolat/mimarsinan/issues/37

dyigitpolat commented 1 year ago

start fixing by never using validation accuracy for target accuracy propagation

dyigitpolat commented 1 year ago

fixed instead by only using validation accuracy.