Abstract: Artificial neural network was introduced into slag glass-ceramic material designing. A 3 layers feedforward network was built with a new robust learning algorithm,
based on a concept of “entire error modifying”. The network has a excellent learning ability when its topology is M-2M-1 and
an appropriate study error chosen. The research results show that this slag glass-ceramic neural network is robust, quick and stable in training
and data predicting, which can disclose the relationship of elemental compositions, structure and material properties of slag glass-ceramic
effectively, even if some parameters are absent in samples.

Key words: artificial neural network, slag glass-ceramic, material designing