Statistical models as building blocks of neural networks

Abstract

The interplay of neural networks and statistical modeling is discussed in the context of the classification problem. It is shown that, on the one hand, the statistical modeling approach provides a systematic way of obtaining good initializations in the neural network context, while, on the other, neural networks offer a powerful expansion to classical model families. A novel integrated approach emerges: statistical models are used as building blocks of neural architectures. The result is an improvement in both flexibility (contribution of neural nets) and interpretability (contribution of statistical modeling).