This paper deals with linear discriminant analysis (LDA) [8] and maximum likelihood linear transformation (MLLT) [9, 10] to estimate the transformation matrices.
This scheme, termed maximum likelihood stochastic transformation (MLST) achieves a good trade-off between robustness and acoustic resolution.
In their systems, adaptation schemes such as feature-space MLLR and maximum likelihood linear transformation (MLLT) greatly contributed to the improvement and achieved about 30%% RWERR.
Adaptive training is performed with constrained maximum likelihood linear regression (CMLLR) transformation [33] of our previously trained average neutral walk HSMM model.
These 44 curves were fitted in three ways: by the classic z-score transformation and least-squares solution, and by probit and logit transformations, using the maximum likelihood method.
Their maximum likelihood method uses an exponential transformation of λ and μ through time with the introduction of two additional parameters, namely k and z, which specify the magnitude of λ decrease and μ increase, respectively: (4a) λ t = λ 0 exp - k t (4b) μ t = μ 0 1 - exp - z t where λ0 is the initial speciation rate, and μ0 the final extinction rate.
Maximum likelihood methods are invariant to linear transformations but, especially in high dimensions, are much more computationally intensive.
SAS Version 9.1 was used for Box-Cox transformations, Newton-Raphson maximum likelihood convergence, pre-diabetic phenotype associations, Fisher's exact test, as well as for all SNP association analyses.
For continuous variables and based on its maximum likelihood estimation we derive a variance-stabilizing transformation and discuss confidence interval construction within the framework of a one-way random effects model.
