HYBRID LOCATION-CONTENT ADDRESSABLE MEMORY WITH ITS APPLICATION TO CHARACTER RECOGNITION
Si-Young Koh, Jeungwon Kim, Bixuan Wang · International Journal of Uncertainty Fuzziness and Knowledge-Based Systems · 1995
This paper reviews the Hybrid Location-Content Addressable Memories (HyLCAMs) and applies them to the printed character recognition problem in order to illustrate their performance in real world problems. Three indirect code generation methods which are the key of the HyLCAM encoding is described to determine a set of proper intermediate states that are linearly separable with inputs. Rejection mechanisms based on code mismatch and rejection region of activation functions are introduced. Computer simulations are performed in order to compare generalization abilities and performance on the rejection mechanism of the HyLCAM models with those of the multilayer perceptrons.