System and method for optimizing pattern recognition of non-gaussian parameters

1. A method comprising:

inputting training data, the training data comprising data for pattern recognition processing;

training an initial objective function having non-Gaussian parameters, the training comprising transforming the initial objective function into, an auxiliary function of parameters that is optimizable computationally more efficiently than the initial objective function;

obtaining parameters that are optimum for the auxiliary function;

obtaining updated parameters by taking a weighted sum of the optimum parameters of the auxiliary function and initial model parameters; and

providing the updated parameters to a pattern recognition system.

2. The method according to claim 1 , wherein the auxiliary function comprises a sum of units that are represented as a composition of modified factors and functional factors.

3. The method according to claim 2 , further comprising transforming said functional factors into countable factors.

4. The method according to claim 2 , wherein said countable factors are represented as quasi-Gaussian factors.

5. The method according to claim 4 , further comprising training the function of the parameter when the function is Gaussian.

6. The method according to claim 2 , wherein said countable factors are represented as polynomials of parameters.

7. The method according to claim 2 , further comprising:

setting the updated model parameters as initial parameters; and

repeating said associating, said obtaining parameters, and said obtaining updated parameters.

8. The method according to claim 1 , further comprising determining whether the function of the parameter is Gaussian.

9. The method according to claim 8 , further comprising training functional factors into quasi-Gaussian factors if the functional factors are not Gaussian.

10. A system comprising:

an input unit configured to input training data, the training data comprising data for pattern recognition processing;

a computer processor comprising:

a training unit configured to train an initial objective function having non-Gaussian parameters, wherein the training comprising transforming the initial objective function into, an auxiliary function of parameters comprising a composition of modified factors and countable factors; and

a unit configured to obtain parameters that are optimum for the auxiliary function; and

a unit configured to obtain updated parameters by taking a weighted sum of the optimum parameters of the auxiliary function and initial model parameters; and

a pattern recognition system configured to receive the updated parameters.

11. The system according to claim 10 , wherein said countable factors are represented as quasi-Gaussian factors.

12. The system according to claim 11 , further comprising a transformation unit for transforming the quasi-Gaussian factors.

13. The system according to claim 12 , wherein training unit that trains the function of the parameter when the function is Gaussian.

14. The system according to claim 10 , further comprising a determination unit that determines whether the function of the parameter is Gaussian.

15. The system according to claim 10 , further comprising a computation unit that computes a function of the composition of modified factors and countable factors.

16. A non-transitory computer-readable storage medium embodying a program of computer-readable instructions executable by a digital processing apparatus to perform a method of optimizing a function of a parameter, said method comprising:

inputting training data, the training data comprising data for pattern recognition processing;

training an initial objective function having non-Gaussian parameters, the training comprising transforming the initial objective function into, an auxiliary function of parameters that is optimizable computationally more efficiently than the initial objective function;

obtaining parameters that are optimum for the auxiliary function;

obtaining updated parameters by taking a weighted sum of the optimum parameters of the auxiliary function and initial model parameters; and

providing the updated parameters to a pattern recognition system.