| Summary: | 碩士 === 國立中山大學 === 應用數學研究所 === 85 === Associative memory is a versatile information processing technique. It plays an important role in pattern recognition tasks. The interpattern associative (IPA) model which enhances the features among the patterns was proved to be one of the most effective method for implementing an associative memory and suitable for both electronic and optical implementation. Two modified IPA models, enhanced interpattern associative (EIPA) model and unipolar enhanced interpattern associative (UEIPA) model, will be proposed in this thesis. Both of them are extended from the original IPA model and can keep more information than the original IPA model. The experimental results show that the performance of both EIPA model and UEIPA model are superior to the original IPA model.
Theoretically, each neuron has its own threshold value. However, in general case, a threshold value, zero, was assigned to all neurons in the network. To improve the network performance, an effective adaptive threshold will also be proposed. Adapting a suitable threshold to each neuron, the network can distinguish the excitatory and inhibitory states of all the stored strings. Combine with the adaptive threshold algorithm for the IPA model, the output state of each neuron in the IPA network can be definitely and correctly determined. The adaptive threshold procedure have improved not only the network performance, but also the storage capacity and noise immunity. The characteristics and limitations of the IPA model are also discussed. Computer simulation and experimental results show that both EIPA model and UEIPA model with adaptive threshold algorithm perform much better than that of without adaptive threshold models.
|
|---|
