A method of using multi-layer perceptrons (MLPs) for modeling complex nonlinear systems is investigated. The importance of pre-processing is crucial to the modeling stage. This includes classifying input/output data i...
详细信息
A method of using multi-layer perceptrons (MLPs) for modeling complex nonlinear systems is investigated. The importance of pre-processing is crucial to the modeling stage. This includes classifying input/output data into different categories for training-data selection, as well as extracting key features of the data. In this paper a prototype problem, an inverted pendulum system, is simulated as a physical system to be identified. The discussion focuses on this problem although the ideas are generic.
A new data processing technique, called the inversion algorithm, is described to produce an undistorted, high resolution and low noise image of tissue. The algorithm is a hybrid approach in which general wave diffract...
详细信息
A new data processing technique, called the inversion algorithm, is described to produce an undistorted, high resolution and low noise image of tissue. The algorithm is a hybrid approach in which general wave diffraction theory is used to extrapolate the propagated part of the acoustic pressure field back to the transducer while a probabilistic method is used to estimate the acoustic propagation velocity.
Some 43,000 lower-limb amputations are performed in the United states each year. Current procedures for fitting a prosthesis to an amputee are somewhat time-consuming and costly, requiring the subjective judgement of ...
详细信息
Some 43,000 lower-limb amputations are performed in the United states each year. Current procedures for fitting a prosthesis to an amputee are somewhat time-consuming and costly, requiring the subjective judgement of trained prosthetists, but necessary to avoid discomfort and ensure successful rehabilitation of the patient. We consider a neural network model which automatically recognizes certain types of misalignment using data obtained from an instrumented shank. Training procedures and partial results are described.
This paper proposes a modeling method of criteria of skilled operators for motion planning of a redundant manipulator in industrial applications. The method employs fuzzy-ID3 to extract important factors with certaint...
详细信息
This paper proposes a modeling method of criteria of skilled operators for motion planning of a redundant manipulator in industrial applications. The method employs fuzzy-ID3 to extract important factors with certainties from the criteria and GMDH (group method of data handling) to model it to evaluate motion plans. Then, it applies a genetic algorithm to optimize redundancy of a manipulator. The proposed method reduces the operator's labor and time for task teaching process thus a path, without considering redundant parameters, only needs to be determined. Experimental results show the effectiveness of the proposed method.
Navigation is one of the standard tasks in the mobile robot domain. In order for a mobile robot to accomplish a non-trivial task, the task should be described in terms of primitive actions of the robot's actuators...
详细信息
Navigation is one of the standard tasks in the mobile robot domain. In order for a mobile robot to accomplish a non-trivial task, the task should be described in terms of primitive actions of the robot's actuators. The complete navigation problem can be broken down into related sub-tasks, which is referred to as "behaviours" in this paper. the authors have used a locally designed mobile robot in an experiment in order to acquire a primitive behaviour, object avoidance. Two different algorithms have been used for comparison purpose. This paper discusses the results of this comparison.
A new classifier based on the Dempster-Shafer theory of evidence is presented. The approach consists in considering the similarity to prototype vectors as evidence supporting certain hypotheses concerning the class me...
详细信息
A new classifier based on the Dempster-Shafer theory of evidence is presented. The approach consists in considering the similarity to prototype vectors as evidence supporting certain hypotheses concerning the class membership of a pattern to be classified. The different items of evidence are represented by basic belief assignments over the set of classes and combined by Dempster's rule of combination. An implementation of this procedure in a neural network with specific architecture and learning procedure is presented. A comparison with LVQ and RBF neural network classifiers is performed.
This paper introduces an improved method for optimizing parameters of an neural network based fuzzy diagnosis module. With the specific structure of a conventional fuzzy system the diagnosis module is used for the lin...
详细信息
This paper introduces an improved method for optimizing parameters of an neural network based fuzzy diagnosis module. With the specific structure of a conventional fuzzy system the diagnosis module is used for the linguistic qualification of continuous signals to detect faulty components in technical processes. The design process of the module structure itself is based on numerical methods applied for neural networks. Training data indicating various system states delivered by a distributed continuous simulator are used to set up the initial module network structure. The proposed multistep parameter learning method enables fast adaptation of the diagnosis module parameters by avoiding mutual influences of parameters during the learning phase and consideration of individual parameter learning characteristics.
This paper introduces a new clustering technique for random data classification based on an enhanced version of the Voronoi diagram. This technique is optimized to deal in the best way possible with data distributions...
详细信息
This paper introduces a new clustering technique for random data classification based on an enhanced version of the Voronoi diagram. This technique is optimized to deal in the best way possible with data distributions which experience overlap in their geometric constructs. A thorough analysis is provided in dealing with the dilemmas imposed by the regions of overlap over the prospect of proper data classification. A mathematical framework is given in view of this enhanced analysis and with respect to the description of real-world data through superposition of Gaussian distributions. Computer results prove the soundness of this clustering technique.
This paper explores shape from motion decomposition as a learning tool for autonomous agents. Shape from motion is a process through which an agent learns the "shape" of some interaction with the world by im...
详细信息
This paper explores shape from motion decomposition as a learning tool for autonomous agents. Shape from motion is a process through which an agent learns the "shape" of some interaction with the world by imparting motion through some subspace of the world. The technique applies singular value decomposition to observations of the motion to extract the eigenvectors. The authors show how shape from motion applied to a fingertip force sensor "learns" a more precise calibration matrix with less effort than traditional least squares approaches. The authors also demonstrate primordial learning on a primitive "infant" mobile robot.
A method to regenerate the control logic automatically from plant operation data is proposed. In this method, the control logic is learned by the inductive learning algorithm. Almost the same control logic as the inst...
详细信息
A method to regenerate the control logic automatically from plant operation data is proposed. In this method, the control logic is learned by the inductive learning algorithm. Almost the same control logic as the installed control logic is obtained in a simulation.
暂无评论