We propose a framework for learning hiddenvariable models by optimizing entropies, in which entropy minimization, posterior maximization, and free energy minimization are all equivalent. Solutions for the maximum a posteriori (MAP) estimator yield powerful learning algorithms that combine all the charms of expectation-maximization and deterministic annealing. Contained as special cases are the ...

The problem addressed in this work is that of separating the voices of simultaneously active moving speakers using a single microphone array. We adapt the multi-sensor multi-target tracking theory to the context of microphone arrays, in order to form a receptive beam that locks on each moving speaker on an extended time basis and, therefore, achieves voice separation. Our approach (a) incorpora...

Predicting the future state of a hybrid system is made exceptionally difficult by uncertainty in the current operating mode of the plant and the possibility of a plant modal transition during the interval over which a prediction is to be made. Unjustified approximations and errors in the system model typically create large errors in the prediction and, for some applications even worse, yield an...

An improved spccd estimator using paraiiicter uncertainty cancellation and its application to an indirect tield-oriented induction motor drive arc presented. First, the speed estimation errors of a V-l model-based adaptive speed observer due to parameter variations are analysed. Then, accordingly, a novel compensation scheme is dcvclopcd to cancel the effects of parameter variations on the esti...

In order to examine dynamic choices under uncertainty, some modelling of individual expectations regarding future variables is necessary. In the context of household mortgage borrowing and saving decisions, particular interest lies in processes for forecasting household income and house prices. Using data from the British Household Panel Survey, the System GMM estimator proposed by Arellano and...

We consider the question of optimally estimating a time-varying multivariate stochastic process, in order to minimize the expected squared estimation error. This is motivated by adaptive control problems under uncertainty in a changing environment. A distinguishing feature of our approach is that we do not need a completely specified model for the stochastic process under consideration. Instead...

Logistic regression (LR) is a famous classification technique commonly used in statistics, machine learning, and data mining area of knowledge for learning a response of binary nature. It assumes that the data values are pre-determined precisely, but this is not true for all conditions. Uncertainty data arises in many applications because of data collection methodology as in repeated measures, ...

A wide variety of processes are thought to show ‘‘long-range persistence’’, specifically an autocorrelation function with power-law decay. A variety of methods have been proposed to quantify this power-law decay, and weather and climate systems, among others, have been claimed to show long-range persistence. In this paper we present a new approach, defining and illustrating a new maximum likeli...

The paper presents a new short-time spectrum estimation algorithm for speech enhancement. A novel multivariate Laplace speech model is utilized to characterize the dependencies between adjacent DFT coefficients of speech, based on which a minimum mean-square error (MMSE) estimator of speech spectral components is derived. Moreover, the speech presence uncertainty is incorporated to modify the M...

In this paper, a modified estimation algorithm has been developed refers to Covariance Shaping Least Square (CSLS) estimation based on the quantum mechanical concepts and constraints. The algorithm has been applied to Auto Regressive Moving Average (ARMA models with various parameter values. The same models can be applied with Colored Noise which estimates the bias in the parameter and the vali...