Abstract
In this paper, we propose novel Cognitive Maps using neural networks named Neural Cognitive Maps (NCMs). Owing to the usage of neural networks, the modeling and prediction abilities of the NCMs are greatly improved compared with those of the conventional FCMs. In order to treat time series inputs effectively, each network receives the latest and the past outputs from all of the networks and is trained by backpropagation algorithm. In addition, a novel pruning method is employed to eliminate useless nodes and weights in each neural network. The pruning has two important effects: one is to improve the performance of the neural network; the other is to give useful information for users to estimate causal relations from the pruned neural network. Computer simulation results indicate the validity and effectiveness of the proposed NCMs.
| Original language | English |
|---|---|
| Pages (from-to) | 3337-3342 |
| Number of pages | 6 |
| Journal | Proceedings of the IEEE International Conference on Systems, Man and Cybernetics |
| Volume | 4 |
| Publication status | Published - 1997 Dec 1 |
| Event | Proceedings of the 1997 IEEE International Conference on Systems, Man, and Cybernetics. Part 3 (of 5) - Orlando, FL, USA Duration: 1997 Oct 12 → 1997 Oct 15 |
ASJC Scopus subject areas
- Control and Systems Engineering
- Hardware and Architecture