Národní úložiště šedé literatury Nalezeno 1 záznamů.  Hledání trvalo 0.00 vteřin. 
Prediktory přesnosti konvolučních neuronových sítí
Karásek, Daniel ; Mrázek, Vojtěch (oponent) ; Piňos, Michal (vedoucí práce)
Last decade has seen a great progress in research of artificial neural network. This progress is mostly consequence of possibility to train larger models than ever before through parallelisation. However researchers reached a point where pure scaling of neural networks does not lead to major improvements. This led to a more complex research of neural network architectures, which introduced new obstacles. The most significant obstacle is the need to evaluate the accuracy of many individual architectures with various hyper-parameters. In some cases even single evaluation can take up to hours on highly specialized computers. One of the methods that can be used to overcome this obstacle is neural network accuracy predictor. Predictors are a group of algorithms that focus on estimating the final validation accuracy of a neural network with no or significantly limited training. This thesis aims to review and reimplement several accuraccy predictors for convolutional neural networks classificators.

Chcete být upozorněni, pokud se objeví nové záznamy odpovídající tomuto dotazu?
Přihlásit se k odběru RSS.