Name: A Review of One-Shot Neural Architecture Search Methods
Journal: Advances in Neural Computation, Machine Learning, and Cognitive Research VI
Authors: Ivan Krivorotov, Egor Maximov, Vladimir Korviakov, Alexey Letunovskiy
Abstract: Neural network architecture design is a challenging and computational expensive problem. For this reason training a one-shot model becomes very popular way to obtain several architectures or find the best according to different requirements without retraining. In this paper we summarize the existing one-shot NAS methods, highlight base concepts and compare considered methods in terms of accuracy, number of needed for training GPU hours and ranking quality.
Link: A Review of One-Shot Neural Architecture Search Methods
Journal: Advances in Neural Computation, Machine Learning, and Cognitive Research VI
Authors: Ivan Krivorotov, Egor Maximov, Vladimir Korviakov, Alexey Letunovskiy
Abstract: Neural network architecture design is a challenging and computational expensive problem. For this reason training a one-shot model becomes very popular way to obtain several architectures or find the best according to different requirements without retraining. In this paper we summarize the existing one-shot NAS methods, highlight base concepts and compare considered methods in terms of accuracy, number of needed for training GPU hours and ranking quality.
Link: A Review of One-Shot Neural Architecture Search Methods