Conformal Risk Control for QoT-Aware Lightpath Feasibility Under Technology-Driven Domain Shift
Organizations: Department of Electrical Engineering, Chalmers University of Technology, 412 96 Gothenburg, Sweden. · University of Applied Sciences and Arts of Southern Switzerland, 6928 Lugano, Switzerland.
Abstract
Artificial intelligence and machine learning (AI/ML)-based regression models support operational decision-making in optical networks but typically expose only point predictions, which carry no statistically controlled information about their reliability. This paper proposes a post-hoc conformal risk control (CRC) framework that keeps a pretrained AI/ML regressor fixed, provides a finite-sample guarantee on uncertainty band miscoverage, and maps the band to operational decisions through policy rules. The guarantee applies to interval miscoverage, while decision-error rates are evaluated empirically. We instantiate the framework for quality of transmission (QoT)-aware lightpath (LP) feasibility assessment, in which a regressor predicts optical signal-to-noise ratio (OSNR) under domain shifts caused by fiber, amplifier, and switching upgrades. Feature shift contribution scores identify the LP features whose relation to QoT changes the most across a technology transition, and two policies translate the bands into feasibility decisions, allowing operators to tune the trade-off between conservatively rejecting or aggressively accepting LP realizations. Experiments on open space-division multiplexing (SDM) QoT datasets over the CONUS topology show that, for a target coverage of 90%, the uncertainty bands meet the target in all 170 trials spanning 17 domain shifts and ten calibration-test partitions, whereas a source-only quantile regression (QR) baseline meets it in none. Under severe shifts, the efficiency-oriented (EO) policy empirically improves zero-shot (ZS) feasibility-decision accuracy from about 85% to above 99%, approaching a transfer learning (TL) policy while avoiding retraining. The framework provides finite-sample control of uncertainty band miscoverage for regression-based optical network decision support, with potential applications beyond QoT-aware LP feasibility assessment.
Figures & tables
| Shift Family | Fiber | Amplifier | Switching | Source Target Scenarios |
|---|---|---|---|---|
| FAS ( FAS ) shift | ✓ | ✓ | ✓ | |
| FA ( FA ) shift | ✓ | ✓ | – | |
| A shift | – | ✓ | – | |
| F shift | ✓ | – | – | |
| S shift (WC-MCF) | – | – | ✓ | |
| S shift (Bu-SMF) | – | – | ✓ |
| LTT uncertainty band | Source-only QR band | |||
|---|---|---|---|---|
| Shift Family | Test cov. (%) | Full width [dB] | Test cov. (%) | Full width [dB] |
| FAS shift | ||||
| FA shift | ||||
| A shift | ||||
| F shift | ||||
| S shift (WC-MCF) | ||||
Appendix figures & tables4 assets
Supplementary material from the paper’s appendix.
Appendix
| Shift family | Source Target | Calibration | Test |
|---|---|---|---|
| FAS ( FAS ) shift | 175,568 | 526,704 | |
| 179,579 | 538,738 | ||
| FA ( FA ) shift | 175,568 | 526,704 | |
| 179,579 | 538,738 | ||
| 180,629 | 541,887 | ||
| A shift | 175,568 | 526,704 |
| LTT uncertainty band | Source-only QR | ||||
| Shift Family | Source Target | Coverage (%) | Width (dB) | Coverage (%) | Width (dB) |
| FAS shift | |||||
| FA shift | |||||
| Calibration samples | |||||||||
|---|---|---|---|---|---|---|---|---|---|
| Shift family | 500 | 1k | 2.5k | 5k | 10k | 25k | 50k | 100k | Full |
| FAS shift | 99.58/4.386 | 97.83/4.290 | 97.94/4.292 | 97.99/4.294 | 97.98/4.293 | 91.69/4.200 | 91.63/4.200 | 91.65/4.200 | 91.65/4.200 |
| FA shift | 99.66/4.345 | 98.06/4.254 | 98.06/4.254 | 98.09/4.254 | 97.87/4.251 | 94.93/4.207 | 91.49/4.161 | 91.50/4.161 | 91.49/4.161 |
| A shift | 99.83/4.822 | 98.42/4.719 | 98.42/4.718 | 98.43/4.719 | 98.40/4.718 | 98.39/4.718 | 96.24/4.687 | 95.74/4.680 | 94.43/4.663 |
| F shift | 99.29/1.086 | 96.82/0.941 | 94.45/0.830 | 93.21/0.775 | 92.52/0.743 | 91.89/0.721 | 91.40/0.707 | 91.17/0.700 | 91.14/0.699 |
| S shift (WC-MCF) | 98.41/0.243 | 96.76/0.229 | 94.48/0.203 | 93.38/0.194 | 92.52/0.188 | 91.80/0.183 | 91.22/0.180 | 91.06/0.179 | 91.03/0.178 |
| Shift family | Cal. samples | Bands selected | RA accuracy | RA FPR | EO accuracy | EO FNR |
|---|---|---|---|---|---|---|
| FAS shift | 500 | 20/20 | – | – | – | – |
| 1k | 20/20 | 70.43 | 0.000 | 99.35 | 0.171 | |
| 2.5k | 20/20 | 70.42 | 0.000 | 99.34 | 0.166 | |
| 5k | 20/20 | 70.41 | 0.000 | 99.34 | 0.163 | |
| 10k | 20/20 | 70.41 | 0.000 | 99.34 | 0.164 | |
| 25k | 20/20 | 70.71 | 0.000 | 99.40 | 0.431 |
Regression Prediction Intervals
Regression prediction intervals quantify plausible ranges for outcomes while balancing reliable coverage against interval width.
See all 119 papers