T. Kohonen, Self-Organizing and Associative Memory. Springer-Verlag, Berling, 3 edition, 1989  Home/Search   Document Not in Database   Summary   Related Articles   Check

....could be eliminated. To fulfil the needs for real time control, the model of the neural networks in this control scheme adopts SOGFCMAC which we present in Section 3 after absorbing the advantages of traditional CMAC [4] fuzzy logic [5] basis functions [6] and SOFM algorithm of Kohonen [7]. In Section 4, we build an actual experimental platform of servo system with low power, on which some experimental researches are done. Finally, we give the conclusion in Section 5. 2. Control Scheme The structure diagram of the neural network online feedback control scheme presented in this ....

....2, M L ]T. The updating rule of weights is: Aw = 0 xTs)W)X, rp (S) 11) g where is the learning rate, and ) is the desired output. lO) vector Because of using the similarity measure based fuzzy addressing scheme, GFCMAC is convenient to be structurally self organized by using SOFM [7]. The GFCMAC after importing SOFM is called SOGFCMAC which is described as follows: Step 1: Initialize: Let L = 0; 1; and epoch=O. Set rp0 (0 rp0 1) max epoch and err. Step 2: Determine the winner of pattern matching: Input s, calculate x and s) by (5) 8) and (10) and choose the ....

Kohonen, T. Self-organizing and associative memory. Springer-Verlag, Berlin, 1988.

....partitionings through self organization and the output subsystem learns responses through the use of eligibility traces. Both input and output learning make use of topological ordering of the neurons and associated neighborhoods, as in Kohonen s Self Organizing Topological Feature Maps [8]. 1.1. Topology and neighborhoods Each SONNET subsystem consists of one or more artificial neural networks. For each network there is a topological ordering of the neurons that remains constant as the network learns. Let D be the dimension of a network. Each neuron is associated with a D tuple ....

....function specifying the width of the neighborhood, then the neighborhood N of neuron n at time t is defined as Nn (t) fu 2 U j d(n; u) W (t)g (1) 1.2. Input space partitioning Input space partitioning in SONNET systems takes place through the self organization principles introduced by Kohonen [8]. For this, a SONNET system may use a single network of dimensionality equal to that of the input space or may use multiple networks of lower dimension, as long as all dimensions of the input space are covered. While it is possible to map a lower dimensional self organizing map onto a ....

[Article contains additional citation context not shown here]

T. K. Kohonen. Self-organizing and associative memory. Springer-Verlag, Berlin, 3rd edition, 1989.

.... input to be stored, say, for further computation) will be stored in weights w(ij) of vector w(j) relating to a j th neuron, if the distance d(ij) satisfies: 2) such storage is known as BAM (Bidirectional Associative Memory) storage [4] Also, a WTA (WinnerTake All) principle is often employed [5], such that an output (firing) is produced only at the winning neuron (say, neuron j, satisfying eqn (2) above) whose weights are closest to vector x(j) when being applied to several neurons during a memory search retrieval task) Another principle, derivable from Hebb s Law [6] and which is ....

....a very large number of categories. The resulting LAMSTAR (LArge Memory STorage And Retrieval) neural network [3,11,12,13] is designed to store and retrieve patterns in a computationally efficient manner, using tools of neural networks, especially SOM (Self Organizing Map) based network modules [5], combined with statistical decision tools. By its structure as described in Section III, the LAMSTAR network is uniquely suited to deal with analytical and non analytical problems [11,12,13] where data are of many vastly different categories and where some categories may be missing, where data ....

[Article contains additional citation context not shown here]

Kohonen, T., (1988): Self-Organizing and Associative Memory, 2 nd Edition, Springer Verlag, N.Y.

....and Liang (1997) treated the two spiral problem with considerable success. Both the 2 25 1 and 2 14 4 1 networks have been tted and the results were close to be perfect, whereas the error rate for BP is generally greater than 40 . In training programs such as back propagation, LVQ algorithms (Kohonen 1989), the total mean squared error E p = X p kO p T p k 2 ; where T p is the p th training case s ideal output and O p is the output of the network, is used as the cost function. We use the same cost function and de ne a probability distribution jointly for the connection strengths w jk and a ....

Kohonen, T. (1989), Self-organizing and Associative Memory, Berlin: Springer-Verlag.

....via a digital transmission line. The system can also be regarded as a basic neural system: the master is a simplified version of integrate fire models [9] 11] and the masters and the slaves are connected based on the winner take all algorithm used in Kohonen s self organizing feature maps [12]. Additionally, in order to obtain not only chaotic pulse trains but also periodic ones, we have applied higher frequency periodic signals to the masters. Response of nonlinear systems to external periodic signals is treated with great interests [13] 15] however, theoretical studies had not ....

T.Kohonen, Self-organizing and associative memory, Springer-Verlag, New York, 1984.

....yet difficult to solve learning control problem. Approaches such as the Cerebellar Model Articulated Controller (CMAC) 25] adaptive fuzzy systems [15] backpropagation through time [18] and fuzzy BOXES [28] have all been applied to versions of this problem. 3. 2 Our Solution Kohonen [14] has proposed a physiologically plausible method of cooperative and competitive organization for connectionist systems that allows them to self organize around a set of input vectors. Several variants on Kohonen s Self Organizing Topological Feature Maps (Kohonen Maps) have been suggested [12] ....

....to a single class for a particular computation, and those outside as belonging to a separate class, giving a discretization which improves the computational efficiency of the method. The concept of the neighborhood relationship is borrowed from Kohonen s SelfOrganizing Topological Feature Maps [14] where the neighborhoods are used for selforganization of the maps. 7 Competition Each neural element in the network is sensitive to a particular region in the input space of the problem. For the present application, the input dimensions are evenly partitioned into eight regions along each ....

T. K. Kohonen. Self-organizing and associative memory. Springer-Verlag, Berlin, 3rd edition, 1989.

....to a single class for a particular computation, and those outside as belonging to a separate class, giving a discretization which improves the computational efficiency of the method. The concept of the neighborhood relationship is borrowed from Kohonen s Self Organizing Topological Feature Maps [4] where the neighborhoods are used for self organization of the maps. 2.2 Competition Each neural element in the network is sensitive to a particular region in the input space of the problem. For the present application, the input dimensions are evenly partitioned into eight regions along each ....

T. K. Kohonen. Self-organizing and associative memory. Springer-Verlag, Berlin, 3rd edition, 1989.

....partitionings through self organization and the output subsystem learns responses through the use of eligibility traces. Both input and output learning make use of topological ordering of the neural elements and associated neighborhoods, as in Kohonen s Self Organizing Topological Feature Maps [8]. 1.1 Topology and neighborhoods Each SONNET subsystem consists of one or more artificial neural networks. For each network there is a topological ordering of the neural elements that remains constant as the network learns. Each neural element is assigned an integer tuple of the same ....

....function specifying the width of the neighborhood, then the neighborhood N of neuron n at time t is defined as Nn (t) fu 2 U j d(n; u) W (t)g (1) 1. 2 Input space partitioning Input space partitioning in SONNET systems takes place through the self organization principles introduced by Kohonen [8]. For this, a SONNET system may use a single network of dimensionality equal to that of the input space or may use multiple networks of lower dimension, as long as all dimensions of the input space are covered. While it is possible to map a lower dimensional self organizing map onto a ....

[Article contains additional citation context not shown here]

T. K. Kohonen. Self-organizing and associative memory. Springer-Verlag, Berlin, 3rd edition, 1989.

....through the use of eligibility traces and adjust their response values on task completion. Response learning can take place rapidly through the use of inter neural coopera tion. This cooperation is based on the neighborhood concept inspired by Kohonen s Self Organizing Topological Feature Maps [8]. ROLNNET systems can be seen as a simplification of the more general learning system introduced by Hougen [4] in which the system also learns the input space partitioning. 2.1 Neighborhood function The internal structure of a ROLNNET map is defined by a topological ordering of the neurons that ....

....to a single class for a particular computation, and those outside as belonging to a separate class, giving a discretization which improves the computational efficiency of the method. The concept of the neighborhood relationship is borrowed from Kohonen s Self Organizing Topological Feature Maps [8] where the neighborhoods are used for self organization of the maps. 2.2 Competition Each neural element in the network is sensitive to a particular region in the input space of the problem. For the present application, in which we have an 8 Theta 8 Theta 8 cubic topology, the three input ....

T. K. Kohonen. Self-organizing and associative memory. Springer-Verlag, Berlin, 3rd ed., 1989.

....all been applied to versions of the trailer backing problem. Our method generally learns faster or with less supervision than these other methods while remaining practical for implementation with minimal computing hardware. Our method combines the self organizing capabilities of Kohonen Maps [ Kohonen, 1989 ] with the temporal sensitivity of eligibility traces. Learning requires constructing a mapping from the input space to output space. The input space is two dimensional. The inputs are (1) the angle from the spine of the trailer to the goal, called the goal angle, and (2) the angle between the cab ....

....of the input space into a number of discrete regions. The sensitivity region weights are initially random within the range of expected input values. As new input values are given to the system, the elements change their sensitivity regions in a self organizing manner, based on Kohonen Maps [ Kohonen, 1989 ] The weights of the selected neural element and of all other elements in its neighborhood are updated to match the input even more closely using w new = w old (t) x w old ) 3) 3 where w is the weight being adjusted, x is the input, and is a time dependent function that determines ....

Kohonen, T. K.: 1989, Self-organizing and associative memory. Berlin: Springer-Verlag, 3rd edition.

....Results section compares the computational results with the human data. The last section discusses brain organization and development in lieu of our results. Similarity in Perception: A window to brain development 6 THE MODEL Model Overview The model is based on a self organizing neural network (Kohonen, 1982, 1989) that is implemented in computer simulations. In order to describe the model in a concrete fashion, we refer to a typical indirect similarity experiment performed by Shepard (1958) This well known experiment used 9 stimuli of distinct red colored chips of uniform size. The colored chips, as ....

.... thus eliminating a scaling bias (see Appendix I for details) The presentation of an input vector gives rise to excitation of neurons in the network array (see Figure 1) The response of neuron r is specified by its n dimensional synaptic weight vector w r and is equal to the dot product of xw r (Kohonen, 1989, 1993) In response to a given input stimulus, the most active neuron in the lattice (for which xw r is maximal) is defined as the winner neuron, s. Its surrounding network activity is modulated by a Gaussian kernel function ( U K 5 centered on neuron s, whose variance R A 2 controls ....

Kohonen, T. (1989). Self-Organizing and Associative Memory. (3rd ed.) Springer, Berlin.

....state codebook is based on a conditional histogram technique that requires storing large conditional block transition probability matrices and extensive computations. In this paper, a Feature Map Finite State VQ (FMFSVQ) is introduced. The FMFSVQ system uses Kohonen s Self Organization algorithm [5] to design a super codebook. This super codebook has nice topological properties that are exploited to simplify the state codebook design. The FMFSVQ system uses the same principles as a Feature Map VQ (FMVQ) designed by [6] except that our encoder and decoder have lower complexity. The DFSVQ, ....

....are physically close together have associated weight vectors that are generally closer in Euclidean distance than neurons that are not physically close. This property will be exploited to simplify our state codebook design. We use a two dimensional Kohenen Self Organization Feature Maps (KSOFM)[5], which has this nice topological property. 3 Feature Map Finite State Vector Quantization 3.1 Encoder Each image is partitioned into small blocks which each block arranged as an n dimensional input vector X k . A super codebook of M codevectors is designed using Kohonen s self organization ....

T. Kohonen, Self-Organizing and Associative Memory, Spring Verlag, 1984.

....to learn is completed. We will refer to these problems as terminal feedback problems. Further, we are interested in problems for which the terminal feedback is no more than a simple boolean value (a success or failure signal) returned by a binary, terminal evaluation function. 3. 2 SONNET Kohonen [12] has proposed a physiologically plausible method of cooperative and competitive organization for connectionist systems that allows them to self organize around a set of input vectors. Several variants on Kohonen s Self Organizing Topological Feature Maps (Kohonen Maps) have been suggested [10] ....

....the computational efficiency of the method. Kohonen Maps combine several simple techniques to achieve powerful self organizing capabilities. The techniques borrowed by SONNET are inter neural competition and intraneighborhood data sharing. For a more complete description of Kohonen Maps, see [12]. 3.3.1 Competition Each neural element in the network has one input weight for each dimension of the input space. Often the input space dimensionality is the same as the dimensionality of the network topology but this is not necessary for some applications. For our example, the input space ....

T. K. Kohonen. Self-organizing and associative memory. Springer-Verlag, Berlin, 3rd edition, 1989.

.... and Liang (1997) treated the two spiral problem with considerable success (both the 2 25 1 and 2 14 4 1 networks have been fitted and the results were close to be perfect, whereas the error rate for BP is generally greater than 40 ) In training programs such as back propagation, LVQ algorithms (Kohonen 1989), the total mean squared error E p = X p kO p Gamma T p k 2 ; where T p is the p th training case s ideal output and O p is the output of the network, is used as the cost function. We use the same cost function and define a probability distribution jointly for the connection strengths w jk ....

Kohonen, T. (1989), Self-organizing and Associative Memory, Berlin: Springer-Verlag.

....been proposed for automatic rule extraction from a given set of input output data examples. For the purpose of extracting fuzzy If Then rules from input output data, we have proposed the Fuzzy Self Organizing Map (FSOM) 2] which has both the architecture of Kohonen s Self Organizing Map (SOM) [1] and the structure of fuzzy If Then rules. In addition, we have also proposed a genetic algorithm using numerical chromosomes and an appropriate crossover method for the numerical chromosomes, called the Unfair Average Crossover [3] We have shown the superiority of our methods to conventional ....

T. Kohonen, "Self-Organizing and Associative Memory." Springer-Verlag, 1989.

....easier we named it BASS (Biopsy Analysis Support System) The system is based on the concept of receptive fields as localized detectors. Receptive fields are biologically inspired [5] 4] In the neural computing literature receptive fields have been used to explain parts of biological vision [7] [8] and other neurophysiological mechanisms. Several computerized approaches to grading of tissue slides are reported in the literature that depend crucially on image segmentation algorithms which generate binary masks. However, these algorithms require a varying degree of skill and attention ....

T. Kohonen, Self-Organizing and associative memory, Springer, Berlin, 1988.

No context found.

T. Kohonen, Self-Organizing and Associative Memory. Springer-Verlag, Berling, 3 edition, 1989

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC