Fig. 3: Performance results from MG and NARMA10 multi-processing.
From: Scalable photonic reservoir computing for parallel machine learning tasks
The figure reports the overlap between the targeted time series (green line) and the predicted output (purple diamonds) for the a MG and b NARMA10 tasks, which were performed in parallel. For each time series, the training results are presented for 300 points used as the validation dataset. We report the highest accuracies achieved with \(g=10.3{{\rm{dB}}}\) (\({V}_{{{\rm{SOA}}}}=1.5\) V) and \({{{\rm{A}}}}_{{{\rm{VOA}}}}\approx 0\) dB (\({V}_{{{\rm{VOA}}}}=0.5{{\rm{V}}}\)) with a 6% NRMSE for the MG task and 1% NRMSE for the NARMA10 task, indicating high classification accuracy for both.
