Abstract: A method for transforming a fuzzy logic system into a neural network, where, in order to simulate membership functions, sigmoid functions are linked together in such a way that, even after the optimization of the neural network, back-transformation of the neural network into a, fuzzy logic system is possible. The advantage of the method described is that a fuzzy logic system can be transformed, in particular component by component, into a neural network and the latter can then be optimized as a whole, i.e. all the components together. The possibility of back-transforming the trained neural network ultimately means that an optimized fuzzy logic system can be obtained. This advantageously makes it possible to use, in particular, standardized fuzzy system software for describing the optimized fuzzy logic system.

Abstract: A neural network for predicting values in non-linear functional mappings having a single hidden layer function generator (12) and an output layer (40). The single hidden layer function generator (12) is operable to receive one or more mapping inputs (x1) and generate a plurality of terms (14) from each mapping input. The plurality of terms generated by the single hidden layer function generator (12) includes at least one trigonometric term selected from the group comprising sin(x1), sin(2x1), sin(3x1), cos(x1), cos(2xl), cos(3xl), cosec(xl), cotan(xl), and being free of Gaussian and Sigmoidal terms.

Abstract: A method of operating a target system wherein small changes in input variables produce small changes in output variables in a manner permits system learning on a time dependent basis. The rate of change of the system output is directly dependent upon the product of the rate of change of the system input and a matrix consisting of input variables and antecedent and consequent parameters. A preferred performance criterion is obtained through the approximation of the said matrix to a weighted-augmented pseudo-inverse Jacobian. Off-line, the system undergoes a series of iterations using a wide range of input variables wherein the actual outputs are compared to the desired outputs and optimized values for the antecedent and consequent parameters are obtained and passed back to the said matrix for use in the subsequent iteration.

Abstract: An artificial intelligence system is provided which makes use of a dual subroutine to adapt weights. Elastic Fuzzy Logic (“ELF”) System is provided in which classical neural network learning techniques are combined with fuzzy logic techniques in order to accomplish artificial intelligence tasks such as pattern recognition, expert cloning and trajectory control. The system may be implemented in a computer provided with multiplier means and storage means for storing a vector of weights to be used as multiplier factors in an apparatus for fuzzy control.

Abstract: A reduced control system suitable for control of a nonlinear or unstable plant is described. The reduced control system is configured to use a reduced sensor set for controlling the plant without significant loss of control quality (accuracy) as compared to an optimal control system with an optimum sensor set. The control system calculates the information content provided by the reduced sensor set as compared to the information content provided by the optimum set. The control system also calculates the difference between the entropy production rate of the plant and the entropy production rate of the controller. A genetic optimizer is used to tune a fuzzy neural network in the reduced controller. A fitness function for the genetic optimizer provides optimum control accuracy in the reduced control system by minimizing the difference in entropy production while maximizing the sensor information content.

Abstract: A self-organizing control system suitable for nonlinear control of a physical object is described. The control system calculates the entropy production difference between a time differentiation (dSu/dt) of the entropy inside the controlled object and a time differentiation (dSc/dt) of the entropy given to the controlled object from a PID controller that controls the object. The entropy production difference is used to generate an evolving control rule by using the entropy production difference as a performance function for a genetic optimizer.

Abstract: A method for transformation of fuzzy logic (FS) into a neural network (NN), in which, in order to form a defuzzified output value (y2) from normalized single-element functions (F1 . . . Fm), the single-element functions (F1 . . . Fm) are each assigned a singleton position (A1 . . . Am) and at least one singleton weighting factor (R1 . . . Rn), those singleton weighting factors (R1 . . . Rn) which are assigned to the same single-element function (F1 . . . Fm) are additively linked, and the singleton weighting factors (R1 . . . Rn) and the additively linked singleton weighting factors (R1 . . . Rn) are weighted via the corresponding singleton positions (A1 . . . Am) and are additively linked in order to form the defuzzified output value (y2). One advantage of the method according to the invention is that the singleton positions (A1 . . .

Abstract: Optimization of a FNN (FNN)-based controller is described. The optimization includes selecting which input signals will be used by the FNN to compute a desired control output. Output parameters are identified and computed by fuzzy reasoning using a neural network. Adjustment of fuzzy rules and/or membership functions for the FNN is provided by a learning process. The learning process includes selecting candidate input data signals (e.g. selecting candidate sensor signals) as inputs for the FNN. The input data is categorized and coded into a chromosome structure for use by a genetic algorithm. The genetic algorithm is used to select an optimum chromosome (individual). The optimum chromosome specifies the number(s) and type(s) of input data signals for the FNN so as to optimize the operation of the FNN-based control system. The optimized FNN-based control system can be used in many control environments, including control of an internal combustion engine.

Abstract: A user preference modeling method using fuzzy networks.

Abstract: A method of processing data relating to a plurality of examples using a data classifier arranged to classify input data into one of a number of classes, and a rule inducer, comprising the steps of: (i) inputting a series of inputs to the data classifier so as to obtain a series of corresponding outputs; (ii) inputting said series of outputs and at least some of said series of inputs to the rule inducer so as to obtain a series of rules which describe relationships between the series of inputs to the data classifier and the series of corresponding outputs from the data classifier.

Abstract: An autonomic system for updating a fuzzy neural network includes a process of calculating an estimated value based on fuzzy inference by using a neural network structure, wherein a parameter to be adjusted or identified by fuzzy inference and outputted from the neural network is made to correspond to coupling loads which are updated by learning, i.e., fuzzy rules and membership functions are adjusted by learning. This system is characterized in that the addition and deletion of fuzzy rules are conducted based on changes in output errors in an autonomic manner, thereby effectively obtaining appropriate numbers of fuzzy rules optimal for an object such as a vehicle engine having strong non-linearity. Fuzzy rules are formed by a combination of membership functions representing variables such as an engine speed and a throttle angle.

Abstract: A set of fuzzy rules (FR) is mapped onto a neural network (NN) (501). The neural network (NN) is trained (502), and weights (wi) and/or neurons (NE) of the neural network (NN) are pruned or grown (503). A new neural network (NNN) formed in this way is mapped onto a new fuzzy rule set (NFR) (504).

Abstract: The subject system provides reduced-dimension mapping of pattern data. Mapping is applied through conventional single-hidden-layer feed-forward neural network with non-linear neurons. According to one aspect of the present invention, the system functions to equalize and orthogonalize lower dimensional output signals by reducing the covariance matrix of the output signals to the form of a diagonal matrix or constant times the identity matrix. The present invention allows for visualization of large bodies of complex multidimensional data in a relatively “topologically correct” low-dimension approximation, to reduce randomness associated with other methods of similar purposes, and to keep the mapping computationally efficient at the same time.

Abstract: A fuzzy logic processor placed in a home automation communications network manages messages traveling through the network. To manage the messages, the fuzzy logic processor works in two stages. In a first stage, on a first part of the message, the fuzzy logic processor ascertains that the message is addressed to it, according to rules by which it recognizes that it is the addressee. Once it recognizes that it is the addressee, it is also capable of producing a routing command because the address of the sender of the signal is present in the message. This routing command is applied to a multiplexer located at the input, interfacing between the fuzzy logic processor and the network in order to route data elements, constituting another part of the message, to the appropriate input of the fuzzy logic processor.

Abstract: A computer modulates among a number of states. Information stored in the computer memory includes predefined categories of an expected external stimulus. For example, the stimulus may include events and the categories include types of events. Also stored in memory are the predetermined states and likelihood functions for transitioning from one state to another. The states may represent emotional states, and the events represent emotion bearing events. Each type of event may have predefined emotional characteristics, with the likelihood functions being response to the occurrence of the events, the categorization of the events, and the characterization of the event's category.

Abstract: An adaptive fuzzy feature mapping (AFFM) technique provides a method for identifying and matching a new data pattern against a set of known data patterns using a combination of distance measurements and fuzzy logic functions. Known data patterns are stored as organized nodes in a pattern map wherein each organized node is defined by one or more attribute coefficients. As distance measurement is computed between a new data pattern and each organized node of the pattern map using distance measurement wherein the organized node having the smallest distance measurement to the new data pattern receives the highest ranking. Traversing the organized nodes according to the ranking, the new data pattern is compared to each organized node using fuzzy logic functions. If the new data pattern matches an organized node based on an acceptable degree of fuzziness, the attribute coefficients of the organized node are updated to reflect those coefficients of the new data pattern.

Abstract: Motor vehicle sensor signals are evaluated by a fuzzy system, which generates control signals for a system device of the motor vehicle—for example an automatic transmission, active suspension, speed stabilization, power-steering assistance, or traction control. The fuzzy system is connected to a neural network, which evaluates the sensor signals and reference data from a recording of driving data of the motor vehicle. The neural network optimizes the rule base of the fuzzy system. During a driving operation, the fuzzy system generates on-line signals categorizing the respective driving situation, and thus makes possible intelligent, time-adaptive, driving-situation-dependent control. The fuzzy system and the neural network each contain a classification system which can be reciprocally converted by a correspondence-maintaining bidirectional transformation.

Abstract: There is disclosed a pattern identifying neural network comprising at least an input and an output layer, the output layer having a plurality of principal nodes, each principal node trained to recognize a different class of patterns, and at least one fuzzy node trained to recognize all classes of patterns recognized by the principal nodes but with outputs set out at levels lower than the corresponding outputs of the principal nodes.

Abstract: Power industry boiler tube failures are a major cause of utility forced outages in the United States, with approximately 41,000 tube failures occurring every year at a cost of $5 billion a year. Accordingly, early tube leak detection and isolation is highly desirable. Early detection allows scheduling of a repair rather than suffering a forced outage, and significantly increases the chance of preventing damage to adjacent tubes. The instant detection scheme starts with identification of boiler tube leak process variables which are divided into universal sensitive variables, local leak sensitive variables, group leak sensitive variables, and subgroup leak sensitive variables, and which may be automatically be obtained using a data driven approach and a leak sensitivity function. One embodiment uses artificial neural networks (ANN) to learn the map between appropriate leak sensitive variables and the leak behavior.

Abstract: An output parameter estimation apparatus for estimating an output parameter from an input data set that is composed of a plurality of input parameters and that is obtained whenever sampling time series input data. In this output parameter estimation apparatus, a fuzzy inference rule is used to calculate a fitness degree of the input data set in one of a plurality of fields included in a space that is formed using at least one input parameter. According to the calculated fitness degree, introduction routes through which the input data set is to be inputted into a neural network are selected. The neural network is set in a connection condition corresponding to the field to which the input data set belongs, the connection condition having been determined in advance as a result of learning process. With this connection condition, the neural network estimates the output parameter from the input data set inputted through the selected introduction routes.

Abstract: In data transfer connections with disturbances, e.g. in cellular networks, it is possible to establish and maintain an optimal connection using the solution according to the present invention. The optimal parameters for each data transfer connection are saved in the modem database. When the user next calls a number, the data of which has been saved, the parameters saved in the database are first used for the connection control. Interference occurring during the data transfer is handled with the help of the database with adaptive logic searching for the optimal data transfer rate and other connection parameters.

Abstract: The invention relates to a modular architecture of a cellular network for improved large-scale integration, of the type which comprises a plurality of fuzzy cellular elements (C.sub.m,n) interconnected to form a matrix of elements having at least m rows and n columns, the row and column numbers describing the location of each element. Each fuzzy processor is adapted for connection to other processors of the same type such that a parallel architecture of the modular type can be implemented. The management of the architecture is facilitated by each submatrix being controlled by an individually dedicated fuzzy processor device.

Abstract: A system is disclosed that provides a goal based learning system utilizing a rule based expert training system to provide a cognitive educational experience. The system provides the user with a simulated environment that presents a business opportunity to understand and solve optimally. Mistakes are noted and remedial educational material presented dynamically to build the necessary skills that a user requires for success in the business endeavor. The system utilizes an artificial intelligence engine driving individualized and dynamic feedback with synchronized video and graphics used to simulate real-world environment and interactions. Multiple "correct" answers are integrated into the learning system to allow individualized learning experiences in which navigation through the system is at a pace controlled by the learner.

Abstract: Method and system for selectively retrieving information contained in a stored document set using a metric-based or "fuzzy" finite-state non-deterministic automation. The system receives a generalized regular search expression from a user. The system then performs prematching during which it estimates a dissimilarity metric for each target string in the stored document set with respect to the search expression. The strings are then sorted by dissimilarity metric, with the best matches, i.e., the strings having the lowest dissimilarity metrics, first. The search expression is broken down into one or more segments. A linear fizzy finite-state non-deterministic automation is constructed (501) by matching each segment of the search expression with a corresponding set of states and transitions. The automation is initialized and then processes target strings read (502) from the sorted list, thereby generating a dissimilarity value for each target string.

Abstract: A method and a system for causing a neural circuit model to learn typical past control results of a process and using the neural circuit model for supporting an operation of the process. The neural circuit model is caused to learn by using, as input signals, a typical pattern of values of input variables at different points in time and, as a teacher signal, its corresponding values of the control variable. An unlearned pattern of input variables is inputted to the thus-learned neuron circuit model, whereby a corresponding value of the control variable is determined. Preferably, plural patterns at given time intervals can be simultaneously used as patterns to be learned.

Abstract: Based on the encoding of deterministic finite-state automata (DFA) in discrete-time, second-order recurrent neural networks, an algorithm constructs an augmented recurrent neural network that encodes a FFA and recognizes a given fuzzy regular language with arbitrary accuracy.

Abstract: An artificial intelligence system is provided which makes use of a dual subroutine to adapt weights. Elastic Fuzzy Logic ("ELF") System is provided in which classical neural network learning techniques are combined with fuzzy logic techniques in order to accomplish artificial intelligence tasks such as pattern recognition, expert cloning and trajectory control. The system may be implemented in a computer provided with multiplier means and storage means for storing a vector of weights to be used as multiplier factors in an apparatus for fuzzy control.

Abstract: A system comprising a neural network, or computer, implementing a feature detection and a statistical procedure, together with fuzzy logic for solving the problem of recognition of faces or other objects) at multiple resolutions is described. A plurality of previously described systems for recognizing faces (or other objects) which use local autocorrelation coefficients and linear discriminant analysis are trained on a data set to recognize facial images each at a particular resolution. In a second training stage, each of the previously described systems is tested on a second training set in which the images presented to the previously described recognition systems have a matching resolution to those of the first training set, the statistical performance of this second training stage being used to train a fuzzy combination technique, that of fuzzy integrals.

Abstract: A fuzzy-neural network system includes: an input layer outputting values of input parameters; a membership layer wherein a multiple number of regions for each of the input parameters are formed by dividing the probable range of the input parameter and a membership function is defined for each of the regions, the membership layer producing membership values as to the regions for each of the input parameters, in accordance with the output values from the input layer; a rule layer wherein specific rules are formed between regions belonging to different input parameters, the rule layer outputting a suitability for each of the rules; an outputting layer producing an output parameter or parameters in accordance with the output values from the rule layer; and a membership value setup means which, if some of the input parameters are unknown, sets up prescribed values as membership values corresponding to the unknown parameters.

Abstract: An inference rule determining process according to the present invention sequentially determines, using a learning function of a neural network model, a membership function representing a degree which the conditions of the IF part of each inference rule is satisfied when input data is received to thereby obtain an optimal inference result without using experience rules. The inventive inference device uses an inference rule of the type "IF . . . THEN . . ." and includes a membership value determiner (1) which includes all of IF part and has a neural network; individual inference quantity determiners (21)-(2r) which correspond to the respective THEN parts of the inference rules and determine the corresponding inference quantities for the inference rules; and a final inference quantity determiner which determines these inference quantities synthetically to obtain the final results of the inference.