pagem@cardiff.ac.uk (Mike Page) (06/04/91)
Has anybody out there done work on masking fields as described in Cohen and Grossberg (1987). I have duplicated the results given in that article, but am finding it difficult to extend these results to larger networks, more specifically networks with larger numbers of masking field populations including some populations which respond to item field sets with mod J greater than three. Parameters for such a network are proving elusive so I'd love to hear from anyone who has already done similar work. Thanks, Mike Page, pagem@uk.ac.cardiff.
marshall@marshall.cs.unc.edu (Jonathan Marshall) (06/05/91)
Mike Page (pagem@cardiff.ac.uk) writes: > Has anybody out there done work on masking fields as described in > Cohen and Grossberg (1987). > > I have duplicated the results given in that article, but am finding > it difficult to extend these results to larger networks, more > specifically networks with larger numbers of masking field populations > including some populations which respond to item field sets with mod J > greater than three. > > Parameters for such a network are proving elusive so I'd love to hear from > anyone who has already done similar work. > > Thanks, > Mike Page, pagem@uk.ac.cardiff. Dear Mike, I've extended Cohen & Grossberg's work on masking fields to allow the networks to self-organize. Some of the advantages of my approach are: o multiple superimposed input patterns can be recognized simultaneously; o uncertainty can be represented by partial activation of classifier neurons; o the networks can be vastly smaller, because not all possible combinations of input patterns need be anticipated; o larger populations can be handled; o analog as well as binary patterns can be handled; o the network can allocate more resources where needed; I use two new learning rules: a Weber-Law adaptive scaling rule, and an inhibitory learning rule. The methods are written up in my paper: J.A. Marshall, "A Self-Organizing Scale-Sensitive Neural Network," Proc. IJCNN, San Diego, June 1990, Vol.III., pp.649-654. Other relevant references are: J.A. Marshall, "Adaptive Neural Methods for Multiplexing Oriented Edges." Intelligent Robots and Computer Vision IX: Neural, Biological, and 3-D Methods, David P. Casasent, Ed. Proceedings of the SPIE 1382, pp.282-291, November 1990. J.A. Marshall, "A Self-Organizing Neural Network for Computing Stereo Disparity and Transparency." Technical Digest, Optical Society of America Annual Meeting, p.268, November 1990. J.A. Marshall, "Representation of Uncertainty in Self-Organizing Neural Networks." Proceedings of the International Neural Network Conference, Paris, France, pp.809-812, July 1990. J.A. Marshall, "Development of Length-Selectivity in Hypercomplex-Type Cells." Investigative Ophthalmology and Visual Science, 31/4, p.397, March 1990. J.A. Marshall, "Self-Organizing Neural Networks for Perception of Visual Motion." Neural Networks, 3, pp.45-74, February 1990. P. Foldiak, (1989). "Adaptive Network for Optimal Linear Feature Extraction." Proceedings of the International Joint Conference on Neural Networks, Washington, DC, June 1989, I., 401-405. P. Foldiak, (1990). "Forming Sparse Representations by Local Anti-Hebbian Learning." Biological Cybernetics, In press. A.L. Nigrin, (1990a). "SONNET: A Self-Organizing Neural Network that Classifies Multiple Patterns Simultaneously." Proceedings of the International Joint Conference on Neural Networks, San Diego, June 1990, II., 313-318. A.L. Nigrin, (1990b). The Stable Learning of Temporal Patterns with an Adaptive Resonance Circuit. Ph.D. Dissertation, Duke University. Please let me know if you hear of any other references or research in this area. Thanks! --Jonathan = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = Jonathan A. Marshall marshall@cs.unc.edu = = Department of Computer Science = = CB 3175, Sitterson Hall = = University of North Carolina Office 919-962-1887 = = Chapel Hill, NC 27599-3175, U.S.A. Fax 919-962-1799 = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =