Fig. 1. Structure of a long-haul WDM transimission optical system^[15]

Fig. 2. Structure of optical fiber channel modelling using FNO^[59]

Fig. 3. Results of optical fiber transmission modeling using FNO^[59]. (a)(c) Time-domain waveforms; (b)(d) frequency-domain waveforms; (e) constellation diagrams

Fig. 4. Block paradigm of fusing physics and machine learning based on PGNN^[60]

Fig. 5. Measured and estimated RSN,NLI of experimental link^[60]

Fig. 6. Diagram of acquiring nonlinear noise by fluctuating optical signal powers^[67]

Fig. 7. Accuracy of EDFA model constructed by active learning and traditional neural networks with randomly-collected dataset under different data size^[67]. (a) Variation of RMSE with data size; (b) violin plot of estimation errors

Fig. 8. EDFA grey-box model^[68]. (a) Flowchart of grey-box modeling scheme; (b) experimental results

Fig. 9. RA grey-box model^[83]. (a) Flowchart of grey-box modeling scheme. (b)(c) experimental results

Fig. 10. Nonlinear monitoring under multicarrier system^[92]. (a) Neural network structure; (b) nonlinear monitoring results for 4-subcarrier system; (c) nonlinear monitoring results for 8-subcarrier system

Fig. 11. Model fusion module for optical filtering penalty monitoring^[93]. (a) Diagram of method; (b) error distribution histogram of baseline scheme using only configuration parameters and fusion model; (c) error distribution histogram of fusion model combining configuration parameters and receiver DSP

Fig. 12. Diagram of acquiring nonlinear noise by fluctuating optical signal powers^[96]

Fig. 13. Error histograms of models on three different links before and after customization^[96]

Fig. 14. Mapping-finding IR paradigm for dynamic optical networks^[109]. (a) Comparison between conventional IR paradigm and mapping-finding IR paradigm; (b)(c) experimental results

Fig. 15. Sensitivity analysis guided memory-aware IR method^[110]. (a) Scheme diagram; (b)(c) experimental results