TFP Measurement and Semi-supervised Isomerous Neural Network (SSINN) with Output Consistency by Interaction

摘要

Total factor productivity (TFP) has aroused great interest in recent years and TFP contribution rate is a very important economic index for a country. Some models such as C-D production function, Malmquist index and data envelopment analysis (DEA) as well as input-out table have been used to compute TFP measurement, but the results are not consistent often. This paper proposes a kind of isomerous neural network with output consistency by interaction. Firstly, we construct three neural networks for TFP measurement with the three models aforesaid respectively. Then the three isomerous models are put the total neural network simultaneity. Their output (TFP) must keep consistency (the square sum of the difference of each other tends to minimum) when the parameters of the models are adjusted respectively. This neural network is neither supervised nor unsupervised, we can call it semi-supervised isomerous NN (SSINN).