Showing papers on "Unsupervised learning published in 1972"

Adaptive pattern recognition: a state-variable approach

Abstract: A state-variable approach to Bayes-optimal adaptive pattern recognition is presented for continuous data systems Both structure and parameter adaptation, as well as supervised and unsupervised learning are considered and Bayes-optimal as well as suboptimal, recursive recognition algorithms are given The state-variable approach consists of modeling random processes involved as the outputs of dynamic systems, linear or nonlinear, excited by white noise, and describing the systems in state-variable form Several fundamental pattern recognition results obtained using the state-variable approach are discussed Specifically, for the class of adaptive pattern recognition problems with signals modeled by nonlinear dynamic systems excited by white gaussian noise and observed in white gaussian noise, the following results are presented and discussed a) The fundamental relationship between pattern recognition and estimation is established Namely, it is shown that pattern recognition/detection constitutes mean-square nonlinear estimation; b) A “partition theorem” is derived that enables decomposition of the nonlinear adaptive pattern recognition system into two parts, a nonadaptive part consisting of recursive matched filters, and an adaptive part that incorporates the learning nature of the adaptive recognition system; c) For the special class of pattern recognition problems with linear dynamic models, the “partition theorem” partitions the nonlinear adaptive recognition system into a linear nonadaptive part consisting of Kalman filters, and a nonlinear adaptive part; d) Several simplified recursive recognition algorithms are presented with substantial computational advantages and high performance; and finally, e) Recursive and computationally efficient algorithms are given for the on-line performance evaluation of the adaptive recognition systems Moreover, two special cases are considered, namely that of supervised learning, treated previously by Lainiotis, and the case of independent signalling random processes For the special case of independent signalling random processes, the results for continuous data are similar to those obtained by Fralick for discrete, conditionally independent data Both deterministic decision-directed learning as well as random decision-directed learning algorithms (Agrawala's LPT) for continuous data are also obtained Moreover, suboptimal, recursive, unsupervised learning algorithms are obtained based on approximate nonlinear estimation procedures

Learning algorithms of pattern recognition in non-stationary conditions

Abstract: Publisher Summary This chapter discusses learning algorithms of pattern recognition in nonstationary conditions. The solution of the problems of learning in pattern recognition under stationary conditions lies in real, iterative algorithms, coinciding in their basis with stochastic approximation type algorithms. The general approach, permitting one to obtain various learning algorithms involves the development of the adaptive approach for learning under nonstationary conditions for learning under stationary conditions. At this stage, the question of recognition under nonstationary conditions is formulated and its solution is found at different stages of information, that is, at different stages of completeness of a priori information. This solution is obtained by considering learning algorithms on the basis of the general approach to learning under nonstationary conditions. The chapter presents a general method for establishing pattern recognition learning algorithms for both supervised and unsupervised learning under nonstationary conditions. The algorithms are learning systems that are capable, with time and after training, of accomplishing recognition of presented patterns.

Information-Theoretic Considerations on Learning

Abstract: The learning procedure in a learning system results in the gradual acquisition of information about the environment. This paper discusses how to measure such information and how it fits into the well-known theory of learning. For this purpose the author introduces a concept of the “learning automaton” that permits the systematic description of most learning machines. It is shown that an expression for the amount of information learned can be determined uniquely up to a constant time under four natural requirements. Apparently, this form is equivalent to the mutual information of information theory. Selecting two simple parametric learning machines as examples, we check the effectiveness of the concept of the amount of learned information in relation to some types of convergence and learning rate.

A learning receiver for communication in three-component multipath channels

Abstract: An adaptive receiver is designed for transmissions through a time-varying multipath channel which may include both specular and diffuse components. The design is based on the theory of unsupervised learning machines and the receiver is a recursive structure which. does not qrow in complex! ty with each new observation, but ii is Bayes' optimal at each instant of time. The multipath mediU!n is modelled as an aggregate of L conditionally independent transmission paths, each consisting of random and/or fixed reflections, and is identified in terms of three components: (1) indirect diffuse scatter, (2) indirect specular reflection, and (3) direct transmission. The channel parameters are time-varying and either independent from one signaling interval to the next or at most M-th order Markov dependent. A review of machines that learn without a teacher is presented and the learning receiver for three-component multipath is designed and modelled on the digital computer. A Monte Carlo simulation is used to estimate the performance when the channel is either Rician or nonfading. This performance, in terms of probability of error, is shown to be consistent with the existing coherent receivers and improves on their performance When the correlation between observations is increased.

Learning characteristics of human operator in man-machine system

Abstract: This paper deals with a problem on the learning characteristics of human operator in a compensatory tracking control task.It is already known that the human operator can learn and adapt himself to changes in control dynamics.It is therefore essential for a system designer to analyze the learning behavior of the human operator. However, the process of learning has not been nivestigated in much detail. A new method to assess the learning process of a human operator is proposed here by fitting the experimental curve to logarithmic function. Variations of the coefficients are then considered to provide a measure for the learning speed.The new method of evaluating the learning process is applied to the cases of step response and frequency response. Also, in order to reduce the training interval, the method of performance scores display is used for the detection of optimal switching point in a relay control system. The result of these investigations presented in this paper are expected to be of use in the design of man-machine systems.