Beyond Feedforward Models Trained by Backpropagation: A Practical Training Tool for a More Efficient Universal Approximator

Ilin, Roman; Kozma, Róbert; Werbos, Paul J.

doi:10.1109/tnn.2008.2000396

Cited by 78 publications

(26 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For Test System I, if the same topology is to be implemented on an MLP, four time-delayed neural networks can be obtained by combining (6) to (9) and replacing the predicted outputs with time delayed values of the actual signal. For example, output of MLP for generator G1 can be obtained using (6), (7) and (9) as follows:…”

Section: ) Performance Comparisonmentioning

confidence: 99%

“…Such a CNN consisting of SRNs as cells are called Cellular SRN (CSRN) and that containing multilayer perceptron (MLP) as cells are called Cellular MLP (CMLP). CSRNs have been used in maze navigation problem [8], [9], facial recognition [10] and image processing [11]. Stability of recurrent neural networks in the presence of noise and time delays is discussed in [12].…”

mentioning

confidence: 99%

See 1 more Smart Citation

Decentralized Asynchronous Learning in Cellular Neural Networks

Luitel

Venayagamoorthy

2012

IEEE Trans. Neural Netw. Learning Syst.

View full text Add to dashboard Cite

Section: ) Performance Comparisonmentioning

confidence: 99%

mentioning

confidence: 99%

Decentralized Asynchronous Learning in Cellular Neural Networks

Luitel

Venayagamoorthy

2012

IEEE Trans. Neural Netw. Learning Syst.

View full text Add to dashboard Cite

“…Within this square matrix, each cell is classified as an obstacle (black), pathway (white), or the goal (red circle). The original CSRN [5] solves a maze by calculating the number of steps to the goal from any cell along the pathway. The network does not solve the exact number of steps towards the goal, but rather solves a maze in such a way that from any pathway cell the nearest neighbor with the lowest value will point in the direction of the shortest path to the goal.…”

Section: A Maze Backgroundmentioning

confidence: 99%

Improved learning in grid-to-grid neural network via clustering

White

Iftekharuddin

Bouzerdoum

2010

The 2010 International Joint Conference on Neural Networks (IJCNN)

View full text Add to dashboard Cite

The maze traversal problem involves finding the shortest distance to the goal from any position in a maze. Such maze solving problems have been an interesting challenge in computational intelligence. Previous work has shown that grid-to-grid neural networks such as the cellular simultaneous recurrent neural network (CSRN) can effectively solve simple maze traversing problems better than other iterative algorithms such as the feedforward multi layer perceptron (MLP). In this work, we investigate improved learning for the CSRN maze solving problem by exploiting relevant information about the maze. We cluster parts of the maze using relevant state information and show an improvement in learning performance. We also study the effect of the number of clusters on the learning rate for the maze solving problem. Furthermore, we investigate a few code optimization techniques to improve the run time efficiency. The outcome of this research may have direct implication in rapid search and recovery, disaster planning and autonomous navigation among others. Abstract-The maze traversal problem involves finding the shortest distance to the goal from any position in a maze. Such maze solving problems have been an interesting challenge in computational intelligence. Previous work has shown that gridto-grid neural networks such as the cellular simultaneous recurrent neural network (CSR ) can effectively solve simple maze traversing problems better than other iterative algorithms such as the feedforward multi layer perceptron (MLP). In this work, we investigate improved learning for the CSR maze solving problem by exploiting relevant information about the maze. We cluster parts of the maze using relevant state information and show an improvement in learning performance. We also study the effect of the number of clusters on the learning rate for the maze solving problem. Furthermore, we investigate a few code optimization techniques to improve the run time efficiency. The outcome of this research may have direct implication in rapid search and recovery, disaster planning and autonomous navigation among others.

show abstract

“…But in order to handle general nonlinear decision problems, the best ADP designs all require several components which learn to approximate unknown nonlinear mappings. Neural networks are essential to handling this task well, in general complex environments, because several types of neural networks provide more accurate universal approximation than any classical alternatives [9,10]. They also make it possible to use new chips in the Cellular Neural Network family which already offer thousands of processors in parallel on a single commercially available chip.…”

Section: The Technology To Meet the Needmentioning

confidence: 99%