Watrous, R. L. "GRADSIM: a connectionist network simulator using gradient optimization techniques," Report, Siemens Corporate Research, Inc., Princeton,New Jersey, 1993  Home/Search   Document Not in Database   Summary   Related Articles   Check

....threshold values are determined by the connection strengths. Figure 5.1 gives a pictorial representation of this example. Note that each connection line contains three links, each containing the indicated delay values. 5. 4 The GRADSIM Neural Network Simulator The GRADSIM neural network simulator [14] was developed by Raymond L. Watrous to provide a connectionist simulator for use in speech recognition research using the temporal flow model. GRADSIM was developed from an early version of the Rochestor Connectionist Simulator [15] and was streamlined for speed providing faster second order ....

Raymond L. Watrous. Gradsim : A connectionist network simulator using gradient optimization techniques. Technical Report MS-CIS-88-16, University of Pennsylvania, March 1988.

....PA 19104 6389 Abstract A data parallel simulator capable of training recurrent time delay connectionist networks is described. The simulator, GRAD CM2, is written in the C programming language and runs on a Connection Machine CM 2. GRAD CM2 is an extension of the serial simulator, GRADSIM [8], and offers similar features. Timing performances of GRAD CM2 and GRADSIM on a series of spatiotemporal discrimination tasks are presented to emphasize the efficacy of data parallelism. 1 Introduction Although the realization that the error gradient of a multi layer network could be computed ....

....(see, for example, 5] A data parallel simulator, capable of training recurrent This work was supported by grant MCS 83 05211 and ARO grants DAA29 84 9 0027 and DAAL03 89C 0031. 1 time delay networks, is described. The simulator, GRAD CM2, is an extension of the serial simulator GRADSIM [8] and offers similar features, including several classical optimization algorithms: fixed step descent, conjugate gradient, and pseudo Newtonian (BFGS) The simulator is written in the C programming language [6] and runs on the Connection Machine model CM 2. Section 2 provides an overview of the ....

[Article contains additional citation context not shown here]

R. Watrous. GRADSIM: a connectionist network simulator using gradient optimization techniques. Technical Report MS-CIS-88-16, University of Pennsylvania, March 1988. 15

....possible to mix and match one with the other. Currently supported are deterministic and stochastic versions of gradient descent, conjugate gradient, quasi Newton s method, and Leverberg Marqardt for search directions, and golden section, cubic interpolation, and a hybrid method for line searches [2, 3]. 4 Neural networks Despite it s name, the neural network module was designed to be extremely general so as to support function approximation schemes that would not commonly be characterized as neural networks. To see how this is possible, we will look in detail at all of the components that ....

.... neural architecture, or calculate the condition number of Table 1: A complete example of using NODElib for learning XOR #include nodelib.h #include stdio.h int testing hook(NN nn) f printf( d d fnn , int)nn x[0] int)nn x[1] nn y[0] return(0) g static double xor data[4][3] = f f 0.0, 0.0, 0.0 g, f 1.0, 0.0, 1.0 g, f 0.0, 1.0, 1.0 g, f 1.0, 1.0, 0.0 g, g; int main(int argc, char argv) f NN nn; Define the neural network. nn = nn create( 2 1 1 ) 2 1 1 architecture. nn link(nn, 0 l 1 ) inputs to hidden link. nn link(nn, 1 l 2 ) ....

R. Watrous. GRADSIM: A connectionist network simulator using gradient optimization techniques. Technical Report NS-CIS-88-16, Dept. of Computer and Information Sciences, University of Pennsylvania, Philadelphia, PA, 1988.

....[38] are used as the soft classifiers [29,36] Networks consist of six input, eight hidden and three output units. The input units are linear, whereas hidden and output units have sigmoid nonlinearities. A conjugate gradient method is used for fast convergence of the supervised learning algorithm [39]. The three outputs correspond to text, image, and non text non image classes. In other words, any sub block not identified as text or image is considered as graphics . Blank regions are detected separately in a straightforward way. The outputs can take values in [0; 1] and the network is trained ....

R.L. Watrous, "GRADSIM: A Connectionist Network Simulator using Gradient Optimization Techniques", TR. MS-CIS-88-16, University of Pennsylvania, March 1988. Submitted to IEEE Transactions on PAMI

No context found.

Watrous, R. L. "GRADSIM: a connectionist network simulator using gradient optimization techniques," Report, Siemens Corporate Research, Inc., Princeton,New Jersey, 1993

No context found.

R.L. Watrous, "GRADSIM: A Connectionist Network Simulator using Gradient Optimization Techniques", TR. MS-CIS-88-16, University of Pennsylvania, March 1988. 30

No context found.

R. L. Watrous. GRADSIM: a connectionist network simulator using gradient optimization techniques. Siemens Corporate Research, Inc., Princeton, New Jersey., 1993.

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