Non-Autoregressive vs Autoregressive Neural Networks for System Identification

The application of neural networks to non-linear dynamic system identification tasks has a long history, which consists mostly autoregressive approaches. Autoregression, the usage model outputs previous time steps, is method transferring state between not necessary for modeling systems with modern network structures, such as gated recurrent units (GRUs) and Temporal Convolutional Networks (TCNs). We compare accuracy execution performance non-autoregressive implementations GRU TCN on simulation task three publicly available benchmarks. Our results show, that are significantly faster at least accurate their counterparts. Comparisons other state-of-the-art black-box methods our implementation best performing network-based method, in benchmarks without extrapolation, method.