高阶神经网络模型特性研究

Research on Higher-Order Neural Networks

研究了高阶神经网络模型阶数和性能之间的关系 .给出了相应的算法和公式 ,并进行了仿真研究 .结果表明 :系统性能的提高并不是简单地与系统阶数成正比 ,对于全互联神经网络模型 ,就存储指标而言 ,只需考虑到四阶相互作用 ,更高阶的考虑对网络性能提高较小而其代价则急剧增大 .

The relationship between the order number and the model property of a higher\|order neural networks was discussed. The algorithm and formulas were presented, and the simulated experiments were carried out. The results show that the improvement of the model characteristics is not proportional to the order of the model. As to the capacity of the neural networks, the 4th\|order interaction is sufficient. Any higher order may bring about a little improvement in property but cost a lot.