Roman Ilin
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Roman Ilin has a total of 6 co-authors. Here are the top co-authors by number of shared publications:
- Paul J. Werbos
Kummer Institute Center for AI and Autonomous Systems, Missouri University of Science and Technology
2 shared publications R. Kozma
Div. of Comput. Sci., Memphis State Univ., TN
2 shared publications- Robert Kozma
University of Memphis
2 shared publications - I. Beliaev
Computer Science Division, The University of Memphis, Memphis, TN USA
1 shared publication P.J. Werbos
1 shared publicationRobert Kozma
1 shared publication
Latest Publications
- 20 May 2008
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks
- Vol. 19 (6) , 929-937
Beyond Feedforward Models Trained by Backpropagation: A Practical Training Tool for a More Efficient Universal Approximator
Cellular simultaneous recurrent neural network (SRN) has been shown to be a function approximator more powerful than the multilayer perceptron (MLP). This means that the complexity of MLP would be prohibitively large for some problems while SRN could realize the desired mapping with acceptable computational constraints. The speed of training of complex recurrent networks is crucial to their successful application. This work improves the previous results by training the network with extended Kalman filter (EKF). We implemented a generic cellular SRN (CSRN) and applied it for solving two challenging problems: 2-D maze navigation and a subset of the connectedness problem. The speed of convergence has been improved by several orders of magnitude in comparison with the earlier results in the case of maze navigation, and superior generalization has been demonstrated in the case of connectedness. The implications of this improvements are discussed.
- 20 May 2008
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks
- Vol. 19 (6) , 929-937
Beyond Feedforward Models Trained by Backpropagation: A Practical Training Tool for a More Efficient Universal Approximator
Cellular simultaneous recurrent neural network (SRN) has been shown to be a function approximator more powerful than the multilayer perceptron (MLP). This means that the complexity of MLP would be prohibitively large for some problems while SRN could realize the desired mapping with acceptable computational constraints. The speed of training of complex recurrent networks is crucial to their successful application. This work improves the previous results by training the network with extended Kalman filter (EKF). We implemented a generic cellular SRN (CSRN) and applied it for solving two challenging problems: 2-D maze navigation and a subset of the connectedness problem. The speed of convergence has been improved by several orders of magnitude in comparison with the earlier results in the case of maze navigation, and superior generalization has been demonstrated in the case of connectedness. The implications of this improvements are discussed.
- 1 August 2007
- Published by Institute of Electrical and Electronics Engineers (IEEE) in 2007 International Joint Conference on Neural Networks
- No. 21614393,p. 2194-2199
Control of multi-stable chaotic neural networks using input constraints
K Sets are nonlinear recurrent connectionist models proposed to emulate the brain dynamics. They can be used as dynamic memories encoding in non-equilibrium attractors. As multidimensional non-linear systems, they are extremely hard to analyze. Their dynamics is strongly believed to be related to the itinerant chaos introduced by Tsuda. In this contribution we design a system with attractor switching based on the previously obtained results. This is a step towards better understanding of the K models and building powerful chaotic neural memory systems.
- 1 August 2007
- Published by Institute of Electrical and Electronics Engineers (IEEE) in 2007 International Joint Conference on Neural Networks
- No. 21614393,p. 2194-2199
Control of multi-stable chaotic neural networks using input constraints
K Sets are nonlinear recurrent connectionist models proposed to emulate the brain dynamics. They can be used as dynamic memories encoding in non-equilibrium attractors. As multidimensional non-linear systems, they are extremely hard to analyze. Their dynamics is strongly believed to be related to the itinerant chaos introduced by Tsuda. In this contribution we design a system with attractor switching based on the previously obtained results. This is a step towards better understanding of the K models and building powerful chaotic neural memory systems.
- 1 April 2007
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 324-329
Efficient Learning in Cellular Simultaneous Recurrent Neural Networks - The Case of Maze Navigation Problem
Cellular simultaneous recurrent neural networks (SRN) show great promise in solving complex function approximation problems. In particular, approximate dynamic programming is an important application area where SRNs have significant potential advantages compared to other approximation methods. Learning in SRNs, however, proved to be a notoriously difficult problem, which prevented their broader use. This paper introduces an extended Kalman filter approach to train SRNs. Using the two-dimensional maze navigation problem as a testbed, we illustrate the operation of the method and demonstrate its benefits in generalization and testing performance
- 1 April 2007
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 324-329
Efficient Learning in Cellular Simultaneous Recurrent Neural Networks - The Case of Maze Navigation Problem
Cellular simultaneous recurrent neural networks (SRN) show great promise in solving complex function approximation problems. In particular, approximate dynamic programming is an important application area where SRNs have significant potential advantages compared to other approximation methods. Learning in SRNs, however, proved to be a notoriously difficult problem, which prevented their broader use. This paper introduces an extended Kalman filter approach to train SRNs. Using the two-dimensional maze navigation problem as a testbed, we illustrate the operation of the method and demonstrate its benefits in generalization and testing performance
- 18 January 2006
- Published by Institute of Electrical and Electronics Engineers (IEEE)
Nonlinear Neurodynamics Tool for System Analysis and Application for Time Series Prediction
Novel type of dynamical neural network, Freeman's K-models has been inspired by the biology and has been studied by researches in a number of research groups. This work is dedicated to providing a unifying test bed implementation that is capable of satisfying needs of possibly the most of these people. Given popularity of MATLABreg computational environment among engineers and scientists our effort is to develop a toolbox that would give way to simulate and experiment with the K-models in a simple manner as well as provide flexible tools for creation of applications that build on this dynamical model. The implementation is presented and detail of the design solutions are provided in this work. Also the successful usage of the toolbox is illustrated with the application to financial time series prediction by the K-models method
- 18 January 2006
- Published by Institute of Electrical and Electronics Engineers (IEEE)
Nonlinear Neurodynamics Tool for System Analysis and Application for Time Series Prediction
Novel type of dynamical neural network, Freeman's K-models has been inspired by the biology and has been studied by researches in a number of research groups. This work is dedicated to providing a unifying test bed implementation that is capable of satisfying needs of possibly the most of these people. Given popularity of MATLABreg computational environment among engineers and scientists our effort is to develop a toolbox that would give way to simulate and experiment with the K-models in a simple manner as well as provide flexible tools for creation of applications that build on this dynamical model. The implementation is presented and detail of the design solutions are provided in this work. Also the successful usage of the toolbox is illustrated with the application to financial time series prediction by the K-models method
- 1 January 2006
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 506-510
Cellular SRN Trained by Extended Kalman Filter Shows Promise for ADP
Cellular simultaneous recurrent neural network has been suggested to be a function approximator more powerful than the MLP's, in particular for solving approximate dynamic programming problems. The 2D maze navigation has been considered as a proof-of-concept task. Present work improves the previous results by training the network with extended Kalman filter (EKF). The original EKF algorithm has been slightly modified. The speed of convergence has been improved by several orders of magnitude in comparison with the earlier results. The implications of this improvement are discussed.
- 1 January 2006
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 506-510
Cellular SRN Trained by Extended Kalman Filter Shows Promise for ADP
Cellular simultaneous recurrent neural network has been suggested to be a function approximator more powerful than the MLP's, in particular for solving approximate dynamic programming problems. The 2D maze navigation has been considered as a proof-of-concept task. Present work improves the previous results by training the network with extended Kalman filter (EKF). The original EKF algorithm has been slightly modified. The speed of convergence has been improved by several orders of magnitude in comparison with the earlier results. The implications of this improvement are discussed.