| 103D2 |
|---|
| Associative Memory by Universal Learning Networks (ULNs) |
| AUTHORS |
| Keiko Shibuta (Kyushu University) Kotaro Hirasawa (Waseda University) Jinglu Hu (Kyushu University) |
| ABSTRACT |
| In this paper, we propose a new auto correlation associative memory using Universal Learning Networks (ULNs). Although so many useful models have been devised, there are some problems related to associative memory, such as the limitation of storage capacity or too small attractors of stored memories. To solve these problems, we obtain a memory network by training network parameters not by calculating them like conventional methods. Furthermore, we introduce ``don't care nodes'' into the networks just to enlarge network size and give more flexibility. We could verify that this method improves the memory capacity by computer simulations. In addition, we studied attractor sizes under the changes of the criterion function used in training parameters. |
Back to Program
Time Table
Top of This Site
Annual Conference 2002