MML Inference of Single-layer Neural Networks

E. Makalic, L. Allison, and D.L. Dowe (Australia)


Architecture selection, Neural Networks, MML, MDL


The architecture selection problem is of great importance when designing neural networks. A network that is too simple does not learn the problem sufficiently well. Con versely, a larger than necessary network presumably in dicates overfitting and provides low generalisation perfor mance. This paper presents a novel architecture selection criterion for single hidden layer feedforward networks. The optimal network size is determined using a version of the Minimum Message Length (MML) inference method. Per formance is demonstrated on several problems and com pared with a Minimum Description Length (MDL) based selection criterion.

Important Links:

Go Back