TRANSFER LEARNING FOR TRAINING ACCELARATION
Ilmārs Apeināns, Vitālijs Žukovs, Sergejs Kodors, Imants Zarembo · HUMAN ENVIRONMENT TECHNOLOGIES Proceedings of the Students International Scientific and Practical Conference · 2021
In this work, authors compare training time of standard convolution neuron network model with model trained using transfer learning. Both models are based on Alexnet architecture. CNN model training from scratch included full model, but using transfer learning, some layers of model were frozen for learning acceleration considering transfer learning methodology.