Long short-term relevance learning

To incorporate sparsity knowledge as well measurement uncertainties in the traditional long short-term memory (LSTM) neural networks, an efficient relevance vector machine algorithm is introduced to network architecture. The proposed scheme automatically determines relevant connections and adapts accordingly, contrast classical LSTM solution. Due its flexibility, new less prone overfitting hence can approximate time-dependent solutions by use of a smaller data set. On structural nonlinear finite element application, we show that self-regulating framework does not require prior suitable architecture size, while ensuring satisfying accuracy at reasonable computational cost.