Combining Architectures for Temporal Learning in Neural-Symbolic Systems

We present a new approach to incorporate a temporal dimension into a hybrid system, by integrating a symbolic model and recurrent neural networks. This combination is supported by an algorithm to perform empirical learning. Further, the network is submitted to testbeds to analyse the influence of background knowledge insertion in the experiments and to validate the algorithm¿s learning capability. Finally, we show that the proposed architecture outperforms existing approaches to temporal learning in connectionist systems.