Fig. 5: Data collection pipeline with minimum convergence sample estimation.
From: Autoencoders for sample size estimation for fully connected neural network classifiers
Stage 1 of the pipeline is to use hyper-parameters to estimate a minimum convergence sample. Stage 2 is to collect the number of samples estimated by the MCSE, and use that to determine the sample size required for a desired performance via MCSE and Eq. (3). Stage 3 is to collect the number of samples required, label those samples and train on the FCN to achieve the desired performance.
