Belangrijkste concepten
Establishing theoretical foundation for ENAS algorithms.
Samenvatting
Evolutionary computation-based neural architecture search (ENAS) algorithms lack theoretical study. This paper proposes a method to estimate the expected hitting time (EHT) of ENAS algorithms, focusing on lower bounds. The process involves common configuration, search space partition, transition probability estimation, population distribution fitting, and hitting time analysis. Theoretical foundation for ENAS algorithms is established through EHT analysis.
Statistieken
Expected hitting time lower bounds are estimated for (λ+λ)-ENAS algorithms with different mutation operators.
Population distribution probabilities are used to calculate the average drift for EHT analysis.
Surface fitting techniques are employed to estimate population distribution based on distance and population size.
Transition probabilities between individuals using various mutation operators are calculated to analyze EHT.
Citaten
"To the best of our knowledge, this work is the first attempt to establish a theoretical foundation for ENAS algorithms."
"Expected hitting time (EHT) signifies the average number of generations needed to find an optimal solution."
"The proposed method integrates theory and experiment for estimating the EHT of ENAS algorithms."