home *** CD-ROM | disk | FTP | other *** search
- This document is divided into two part as follows.
-
- 1. Questions About This Package
- 2. Questions About The Theory Behind This Package
-
-
-
- 1. Questions About This Package
-
- Q; What capabilities does this software package have that
- differentiate it from those developed elsewhere ?
- A; This package
- (1) Includes a network structure estimation program, that allows
- one to estimate how many hidden units an MLP must have to achieve
- a user-chosen performance. This program usually, but not always, works.
- (2) Includes a fast training program unlike others that are available.
- This technique is about 3 times faster than full-blown (without
- heuristic changes for speeding it up) conjugate gradient training,
- and performs slightly better. Training is 10 to 100 times faster
- than backpropagation.
- (3) Includes a network structure analysis program. Given a trained
- MLP, this program makes a table of network performance versus
- the number of hidden units. Using the table, and the non-demo
- version of this program, the user can choose the size wanted
- and prune the network, saving new weight and network structure files.
- The non-demo version can also determine the amount of nonlinearity
- (degree) in each hidden layer, thereby informing the user if a
- linear network would solve his problem.
-
- Q; Why does this package design only classification networks and not
- mapping or estimation networks ? Don't both types of network use
- the same format for training data and the same training algorithms ?
-
- A; We have separate packages for classification and mapping or
- estimation because;
- (1) Our training algorithms for classification and mapping networks have
- some important differences. For example, the functional link net
- design for a mapping net is not iterative, whereas that for
- classification nets is iterative. The MLP classification network
- learns even when the learning factor is set to 0., unlike the MLP
- for mapping.
- (2) Combining the two packages would make the result unnecessarily large.
- (3) Many people need to do mapping or classification but not both.
-
- Q; What error function is being minimized during backpropagation training,
- fast training, and functional link net training ?
-
- Nout
- A; MSE = (1/Npat) SUM MSE(k) where
- k=1
-
- Npat 2
- MSE(k) = SUM [ Tpk - Opk ]
- p=1
-
- where Npat is the number of training patterns, Nout is the number
- of network output nodes, Tpk is the desired output for the pth
- training pattern and the kth output, and Opk is the actual output
- for the pth training pattern and the kth output. MSE is printed
- for each iteration.
-
- Q; What are "RMS error", "Relative RMS Error" , and "Error Variances"?
- The rms error of the kth output, RMS(k), is SQRT( MSE(k)/Npat ),
- where SQRT means square root. The kth output's Relative RMS Error is
-
- R(k) = SQRT( MSE(k)/E(k) ) where
-
- Npat 2
- E(k) = SUM [ Opk-Mk ] and
- p=1
-
- Npat
- Mk = (1/Npat) SUM Opk
- p=1
-
- The kth output's Error Variance is MSE(k)/Npat.
-
-
- Q; I get an "Out of environment space when deleting a file in the
- utilities section. How can I fix this ?
- A; Increase the environment space. Add the switch /e:512 or /e:1024
- to the shell command in your config.sys file.
-
- Q: The package seems very sluggish when I use a serial mouse. How can
- I fix this ?
- A; The package should run much faster if you disable the mouse. You can
- comment out the mouse command in your autoexec.bat file. Also, a
- bus type mouse may work ok. We are still investigating the serial
- mouse problem.
-
-
- 2. Questions About The Theory Behind This Package
-
- Q; Do you have any papers related to the prediction of neural net
- size (Sizing) ?
- A; Classified.
-
-
- Q; Do you have any papers related to fast training of MLPs, and
- related topics?
- A; Yes.
-
- M.S. Dawson, A.K. Fung, M.T. Manry, "Sea Ice Classification Using
- Fast Learning Neural Networks," Proc. of IGARSS'92, Houston, Texas,
- May 1992, vol. II, pp 1070-1071.
-
- M.S. Dawson, J. Olvera, A.K. Fung, M.T. Manry, "Inversion of
- Surface Parameters Using Fast Learning Neural Networks," Proc. of
- IGARSS'92, Houston, Texas, May 1992, vol. II, pp 910-912.
-
- M.T. Manry, X. Guan, S.J. Apollo, L.S. Allen, W.D. Lyle, and W.
- Gong, "Output Weight Optimization for the Multi-Layer Perceptron,"
- Conference Record of the Twenty-Sixth Annual Asilomar Conference on
- Signals, Systems, and Computers, Oct. 1992, vol 1, pp. 502-506.
-
- X. Jiang, Mu-Song Chen, and M.T. Manry, "Compact Polynomial
- Modeling of the Multi-Layer Perceptron," Conference Record of the
- Twenty-Sixth Annual Asilomar Conference on Signals, Systems, and
- Computers, Oct. 1992, vol 2, pp.791-795.
-
- R.R. Bailey, E.J. Pettit, R.T. Borochoff, M.T. Manry, and X. Jiang,
- "Automatic Recognition of USGS Land Use/Cover Categories Using
- Statistical and Neural Network Classifiers," Proceedings of SPIE
- OE/Aerospace and Remote Sensing, April 12-16, 1993, Orlando
- Florida.
-
- M.S. Dawson, A.K. Fung, M.T. Manry, "Classification of SSM/I Polar
- Sea Ice Data Using Neural Networks," Proc. of PIERS 93, 1993, p.
- 572.
-
- F. Amar, M.S. Dawson, A.K. Fung, M.T. Manry, "Analysis of
- Scattering and Inversion From Forest," Proc. of PIERS 93, 1993, p.
- 162.
-
- A. Gopalakrishnan, X. Jiang, M-S Chen, and M.T. Manry,
- "Constructive Proof of Efficient Pattern storage in the Multilayer
- Perceptron," Conference Record of the Twenty-Seventh Annual
- Asilomar Conference on Signals, Systems, and Computers, Nov. 1993.
-
- K. Rohani, M.S. Chen and M.T. Manry, "Neural Subnet Design by
- Direct Polynomial Mapping," IEEE Transactions on Neural Networks,
- Vol. 3, no. 6, pp. 1024-1026, November 1992.
-
- M.S. Dawson, A.K. Fung, and M.T. Manry, "Surface Parameter
- Retrieval Using Fast Learning Neural Networks," Remote Sensing
- Reviews, Vol. 7, pp. 1-18, 1993.
-
- M.T. Manry, S.J. Apollo, L.S. Allen, W.D. Lyle, W. Gong, M.S.
- Dawson, and A.K. Fung, "Fast Training of Neural Networks for Remote
- Sensing," Remote Sensing Reviews, July 1994, vol. 9, pp. 77-96, 1994.
-
- Q; Do you have any papers related to the analysis of trained neural
- networks ?
- A; Yes.
-
- W. Gong and M.T. Manry, "Analysis of Non-Gaussian Data Using a
- Neural Network," Proceedings of IJCNN 89, vol. II, p. II-576,
- Washington D.C., June 1989.
-
- M.S. Chen and M.T. Manry, "Back-Propagation Representation Theorem
- Using Power Series," Proceedings of IJCNN 90, San Diego, I-643 to
- I-648.
-
- M.S. Chen and M.T. Manry, "Basis Vector Analyses of Back-
- Propagation Neural Networks," Proceedings of the 34th Midwest
- Symposium on Circuits and Systems, Monterey, California, May 14-17
- 1991, vol. 1, pp 23-26.
-
- M.S. Chen and M.T. Manry, "Power Series Analyses of Back-
- Propagation Neural Networks," Proc. of IJCNN 91, Seattle WA., pp.
- I-295 to I-300.
-
- M.S. Chen and M.T. Manry, "Nonlinear Modelling of Back- Propagation
- Neural Networks," Proc. of IJCNN 91, Seattle WA., p. A-899.
-
- M.S. Chen and M.T. Manry, "Basis Vector Representation of Multi-
- Layer Perceptron Neural Networks," submitted to IEEE Transactions
- on Neural Networks.
-
- W. Gong, H.C. Yau, and M.T. Manry, "Non-Gaussian Feature Analyses
- Using a Neural Network," accepted by Progress in Neural Networks,
- vol. 2, 1991.
-
- X. Jiang, Mu-Song Chen, M.T. Manry, M.S. Dawson, A.K. Fung,
- "Analysis and Optimization of Neural Networks for Remote Sensing,"
- Remote Sensing Reviews, July 1994, vol. 9, pp. 97-114, 1994.
-
- M.S. Chen and M.T. Manry, "Conventional Modelling of the Multi-
- Layer Perceptron Using Polynomial Basis Functions," IEEE
- Transactions on Neural Networks, Vol. 4, no. 1, pp. 164-166,
- January 1993.
-
- K. Rohani and M.T. Manry, "Multi-Layer Neural Network Design Based
- on a Modular Concept," accepted by the Journal of Artificial Neural
- Networks.
-
- Q; Do you have any papers related to the prediction of neural net
- performance, and pre-processing of data ?
- A; Yes.
-
- S.J. Apollo, M.T. Manry, L.S. Allen, and W.D. Lyle, "Optimality of
- Transforms for Parameter Estimation," Conference Record of the
- Twenty-Sixth Annual Asilomar Conference on Signals, Systems, and
- Computers, Oct. 1992, vol. 1, pp. 294-298.
-
- Q. Yu, S.J. Apollo, and M.T. Manry, "MAP Estimation and the
- Multilayer Perceptron," Proceedings of the 1993 IEEE Workshop on
- Neural Networks for Signal Processing, Linthicum Heights, Maryland,
- Sept. 6-9, 1993, pp. 30-39.
-
- S.J. Apollo, M.T. Manry, L.S. Allen, and W.D. Lyle, "Theory of
- Neural Network-Based Parameter Estimation," submitted to Neural
- Network Trans. of the IEEE.
-
- S.J. Apollo, M.T. Manry, L.S. Allen, and W.D. Lyle,
- "Transformation-Based Data Compression for Parameter Estimation,"
- submitted to IEEE Trans. on Signal Processing.
-
-
-
- Q; Do you have any papers related to the training of functional link
- neural networks ?
- A; Yes.
-
- H.C. Yau and M.T. Manry, "Sigma-Pi Implementation of a Nearest
- Neighbor Classifier," Proceedings of IJCNN 90, San Diego, I-667 to
- I-672.
-
- H.C. Yau and M.T. Manry, "Sigma-Pi Implementation of a Gaussian
- Classifier," Proceedings of IJCNN 90, San Diego, III-825 to
- III-830.
-
- H.C. Yau and M.T. Manry, "Shape Recognition Using Sigma-Pi Neural
- Networks," Proc. of IJCNN 91, Seattle WA., p. II A-934.
-
- H.C. Yau and M.T. Manry, "Shape Recognition with Nearest Neighbor
- Isomorphic Network," Proceedings of the First IEEE-SP Workshop on
- Neural Networks for Signal Processing, Princeton, New Jersey, Sept.
- 29 - Oct. 2, 1991, pp. 246-255.
-
- H.C. Yau and M.T. Manry, "Iterative Improvement of a Gaussian
- Classifier," Neural Networks, Vol. 3, pp. 437-443, July 1990.
-
- H.C. Yau and M.T. Manry, "Iterative Improvement of a Nearest
- Neighbor Classifier," Neural Networks, Vol. 4, Number 4, pp.
- 517-524, 1991.
-