Title
Feature based retrieval in neural networks
Document Type
Conference Proceeding
Publication Date
12-1-1992
Abstract
A model of a neural system is presented that achieves feature based retrieval. It establishes the principle that any final state of the system can be uniquely determined by a subset of initially active neurons provided this uniqueness is preserved throughout the evaluation of the quiescent neurons. A method that finds distinguishing features chooses the initial neurons. Loosely ordering the spread of activation preserves uniqueness by maintaining the information conveyed by the growing active set. Finally, proper ordering is achieved under semi-random and asynchronous conditions using a function implemented locally by each neuron that minimizes the energy a newly activated neuron contributes. This is significant if such a system is fabricated since it avoids global control.
Publication Source (Journal or Book title)
Conference Proceedings - IEEE SOUTHEASTCON
First Page
819
Last Page
821
Recommended Citation
Young, D., & Kak, S. (1992). Feature based retrieval in neural networks. Conference Proceedings - IEEE SOUTHEASTCON, 2, 819-821. Retrieved from https://repository.lsu.edu/physics_astronomy_pubs/6021