English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the zaia usage as premium feature

ZAIA

Presents the pdf analysis as premium feature

PDF Analysis

Insights

Presents the summarisation as premium feature

Summarisation

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the access of premium content as premium feature

Premium Content

Zendy Tools

Zendy Open

Neural network models for dynamic systems can be trained either in parallelor in series-parallel configurations. Influenced by early arguments, severalpapers justify the choice of series-parallel rather than parallel configurationclaiming it has a lower computational cost, better stability properties duringtraining and provides more accurate results. Other published results, on theother hand, defend parallel training as being more robust and capable ofyielding more accu- rate long-term predictions. The main contribution of thispaper is to present a study comparing both methods under the same unifiedframework. We focus on three aspects: i) robustness of the estimation in thepresence of noise; ii) computational cost; and, iii) convergence. A unifyingmathematical framework and simulation studies show situations where eachtraining method provides better validation results, being parallel trainingbetter in what is believed to be more realistic scenarios. An example usingmeasured data seems to reinforce such claim. We also show, with a novelcomplexity analysis and numerical examples, that both methods have similarcomputational cost, being series series-parallel training, however, moreamenable to parallelization. Some informal discussion about stability andconvergence properties is presented and explored in the examples.

"Parallel Training Considered Harmful?": Comparing series-parallel and parallel feedforward network training