Mixture Density Network Estimation of Continuous Variable Maximum Likelihood Using Discrete Training Samples
This work addresses performance issues in statistical estimation for applications where only discrete training data is available, representing an incremental improvement.
The paper tackles the problem of biased parameter estimates when using Mixture Density Networks (MDNs) with discrete training samples for continuous variable maximum likelihood, and proposes corrective methods to address these biases.
Mixture Density Networks (MDNs) can be used to generate probability density functions of model parameters $\boldsymbolθ$ given a set of observables $\mathbf{x}$. In some applications, training data are available only for discrete values of a continuous parameter $\boldsymbolθ$. In such situations a number of performance-limiting issues arise which can result in biased estimates. We demonstrate the usage of MDNs for parameter estimation, discuss the origins of the biases, and propose a corrective method for each issue.