Comparison with Offset-Based and Quantile-Regression Reconfiguration Triggers

Compare the Mixture Density Network-based proactive controller against deterministic reconfiguration baselines using fixed safety offsets and against triggers based on quantile regression for dynamic rerouting under optical span degradation.

Background

The dynamic reconfiguration experiment compares a reactive multilayer perceptron trigger, which reroutes when the predicted Q-margin falls below 0 dB, with a Mixture Density Network trigger that reroutes when the estimated violation probability exceeds a risk threshold. Although the probabilistic controller reduces failures by acting before the predicted margin crosses the feasibility boundary, the evaluation does not compare it with deterministic baselines that trigger at a fixed positive safety offset, nor with quantile-regression-based triggers. Such comparisons would clarify whether the observed benefits arise specifically from modeling a full conditional distribution or can also be obtained through simpler uncertainty or margin rules.