Skip to main content
Fig. 2 | BMC Bioinformatics

Fig. 2

From: Mixture density networks for the indirect estimation of reference intervals

Fig. 2

Component switching. Identifiability problems promoted by unfavorable random initial model weights illustrated by the estimated location parameters alternating between the true component means. Dark grey lines represent the main component, with the slightly more wide solid lines representing the true location parameter of the data-generating process. Left: Results from the BFGS optimizer on the fully connected MDN. Right: Results of the ADAM optimizer with early stopping

Back to article page