Abstract: A method to obtain the Pareto solutions that are specified by human preferences is suggested. The main idea is to convert the fuzzy preferences into interval-based weights. With the help of the dynamically-weighted aggregation method, it is shown to be successful to find the preferred solutions on two test functions with a convex Pareto front. Compared to the method described in “Use of Preferences for GA-based Multi-Objective Optimization” (Proceedings of 1999 Genetic and Evolutionary Computation Conference, pp. 1504-1510, 1999) by Cvetkovic et al., the method according to the invention is able to find a number of solutions instead of only one, given a set of fuzzy preferences over different objectives. This is consistent with the motivation of fuzzy logic.

Abstract: In the field of multi-objective optimization using evolutionary algorithms conventionally different objectives are aggregated and combined into one objective function using a fixed weight when more than one objective needs to be optimized. With such a weighted aggregation, only one solution can be obtained in one run. Therefore, according to the present invention two methods to change the weights systematically and dynamically during the evolutionary optimization are proposed. One method is to assign uniformly distributed weight to each individual in the population of the evolutionary algorithm. The other method is to change the weight periodically when the evolution proceeds. In this way a full set of Pareto solutions can be obtained in one single run.

Abstract: The invention relates to an evolutionary optimization method. First, an initial population of individuals is set up and an original fitness function is applied. Then the offspring individuals having a high evaluated quality value as parents are selected. In a third step, the parents are reproduced to create a plurality of offspring individuals. The quality of the offspring individuals is evaluated selectively using an original fitness function or an approximate fitness function. Finally, the method returns to the selection step until a termination condition is met. The step of evaluating the quality of the offspring individuals includes grouping all offspring individuals in clusters, selecting for each cluster one or a plurality of offspring individuals, resulting in altogether selected offspring individuals, evaluating the selected offspring individuals by the original fitness function, and evaluating the remaining offspring individuals by means of the approximate fitness function.

Abstract: For object recognition, an image is segmented into areas of similar homogeneity at a coarse scale, which are then interpreted as surfaces. Information from different spatial scales and different image features is simultaneously evaluated by exploiting statistical dependencies on their joint appearance. Thereby, the local standard deviation of specific gray levels in the close environment of an observed pixel serves as a measure for local image homogeneity that is used to get an estimate of dominant global object contours. This information is then used to mask the original image. Thus, a fine-detailed edge detection is only applied to those parts of an image where global contours exist. After that, said edges are subject to an orientation detection. Moreover, noise and small details can be suppressed, thereby contributing to the robustness of object recognition.

Abstract: A method and system for obtaining a pool of speech syllable models. The model pool is generated by first detecting a training segment using unsupervised speech segmentation or speech unit spotting. If the model pool is empty, a first speech syllable model is trained and added to the model pool. If the model pool is not empty, an existing model is determined from the model pool that best matches the training segment. Then the existing module is scored for the training segment. If the score is less than a predefined threshold, a new model for the training segment is created and added to the pool. If the score equals the threshold or is larger than the threshold, the training segment is used to improve or to re-estimate the model.

Abstract: The invention relates to the field of automated processing of speech signals and particularly to a method for tracking the formant frequencies in a speech signal, comprising the steps of: obtaining an auditory image of the speech signal; sequentially estimating formant locations; segmenting the frequency range into sub-regions; smoothing the obtained component filtering distributions; and calculating exact formant locations.

Abstract: A method for modifying an object design using a computer comprises the steps of: selecting a first sub-design of the object design comprising a first free form deformation geometry and a first free deformation control volume that is variable and adaptive; choosing a second sub-design comprising a second free form deformation geometry; and replacing the first geometry with the second geometry.

Abstract: Controlling an orienting/positioning system having a sensor and actuator for controlling at least one of an orienting and positioning action of the sensor. The invention (1) evaluates pre-action output information of a sensor in order to detect the position of a pattern in the input space of the sensor, (2) determines a targeted post-action position of the pattern in the input space of the sensor, (3) defines an actuator command by mapping any deviation of the pre-action and post action position in the input space to actuator control coordinates using a predefined mapping function, (4) controls the actuators according to the defined command to execute the orienting/positioning action, (5) detects the real post-action position of the pattern in an input space of the sensor, (6) adapts the mapping function based on differences between the real post-action position and the targeted post-action position of the pattern in the input space.

Abstract: A method for controlling behavior of an intelligent mechanical system having more than one degree of freedom, the method comprising: (a) providing a target to a selector unit that requires operation of actuators of the mechanical system; (b) generating by a selector unit more than one behavior commands adapted to reach the target; (c) simulating movements of the mechanical system for each behavior command by a computing unit of the mechanical system; (d) assessing fitness values for each simulated movements based on at least one objective; and (e) sending the behavior commands with a highest fitness value to the actuators of the mechanical system.

Abstract: An interactive robot comprises visual sensors, manipulators and a computer. The computer is configured to generate proto-objects from output signals from the visual sensors and store the proto-objects in memory of the computer. The proto-objects represent blobs of interest in an input field of the visual sensors identified at least by a three dimensional position label. The computer also generates objects hypotheses representing a category of the object based on evaluation of the proto-objects with respect to different behavior-specific constraints; and determines a visual tracking movement of the visual sensor, a movement of a body of the robot or a movement of the manipulators based on the object hypotheses and at least one proto-object as a target.

Abstract: An artificial ear for a robot comprises an auricle-shaped structure, a microphone which in a concha region of the auricle-shaped structure, wherein the lower or rear edge of the auricle-shaped structure forms acoustic shield for shadowing noise from any noise source located below or behind the concha region.

Abstract: A method for estimating the location of a sound source. First, a first microphone and a second microphone are moved relative to the sound source. Then the Intra-aural Time Difference (ITD) between the two microphones is measured. The moving and the measuring are repeated until the Intra-aural Time Difference (ITD) is smaller than a predefined threshold.

Abstract: Within the frameworks of hierarchical neural feed-forward architectures for performing real-world 3D invariant object recognition a technique is proposed that shares components like weight-sharing (2), and pooling stages (3, 5) with earlier approaches, but focuses on new methods for determining optimal feature-detecting units in intermediate stages (4) of the hierarchical network. A new approach for training the hierarchical network is proposed which uses statistical means for (incrementally) learning new feature detection stages and significantly reduces the training effort for complex pattern recognition tasks, compared to the prior art. The incremental learning is based on detecting increasingly statistically independent features in higher stages of the processing hierarchy. Since this learning is unsupervised, no teacher signal is necessary and the recognition architecture can be pre-structured for a certain recognition scenario.

Abstract: In a method for controlling an orienting/positioning system, the orienting system comprises at least one sensor means, an external motion detection device for detecting an externally-caused motion in the environment of the sensor means and actuator means for controlling an orienting and/or positioning action of the sensor means.

Abstract: A system and method for combining the model-based and genetics-based methods are combined according to a convergence criterion. When the population is not converged, the genetics-based approach is used, and when the population is converged, the model-based method is used to generate offspring. The algorithm benefits from using a model-based offspring generation only when the population shows a certain degree of regularity, i.e., converged in a stochastic sense. In addition, a more sophisticated method to construct the stochastic part of the model can be used. Also a biased Gaussian noise (the mean of the noise is not zero), as well as a white Gaussian noise (the mean of the noise is zero) can be preferably used for the stochastic part of the model.

Abstract: The present invention relates to a optimization method based on an evolution strategy according to which a model/structure/shape/design to be optimized is described by parameter sets comprising object parameters. The object parameters are mutated to create offsprings of the parameter set. The quality of the offsprings is evaluated. The parameter set furthermore comprises at least one strategy parameter representing the step-size of the mutation (f.e. the variance of the normal distribution) of associated object parameters. The number of object parameters as well as the number of associated strategy parameters can be adapted during the optimization process. The value of newly inserted strategy parameters can be estimated based on the information of correlated object parameters.

Abstract: Convolutional networks can be defined by a set of layers being respectively made up by a two-dimensional lattice of neurons. Each layer—with the exception of the last layer—represents a source layer for respectively following target layer. A plurality of neurons of a source layer called a source sub-area respectively share the identical connectivity weight matrix type. Each connectivity weight matrix type is represented by a scalar product of an encoding filter and a decoding filter. For each source layer a source reconstruction image is calculated on the basis of the corresponding encoding filters and the activities of the corresponding source sub-area. For each connectivity weight matrix type, each target sub-area and each target layer the input of the target layer is calculated as a convolution of the source reconstruction image and the decoding filter.

Abstract: A framework for managing approximate models in generation-based evolution control is proposed. This framework is well suited for parallel evolutionary optimization that is able to guarantee the correct convergence of the evolutionary algorithm and to reduce the computation costs as much as possible. Control of the evolution and updating of the approximate models are based on the estimated fidelity of the approximate model. The frequency at which the original function is called and the approximate model is updated is determined by the local fidelity of the approximate model. By local fidelity the fidelity of the model for the region where the current population is located is designated. The lower the model fidelity is, the more frequently the original function should be called and the approximate models should be updated.