Learning Not to Optimize: Physics-Informed Action-Space Reshaping for Intent-Based Network Control

2026-08-01

AI Summary

This work addresses the inefficiency of conventional policy optimization in intent-based network control, where excessive exploration of ineffective actions persists before value function convergence. To overcome this limitation, the authors propose the "Learn Not to Optimize" (LNOQRD) framework, which dynamically prunes equivalent, suboptimal, or infeasible actions through quotient space modeling under state-intent relabeling, dominance relation analysis, and physics-informed residual filtering. By integrating group equivariance, monotonicity theory, and approximate certificate reasoning, LNOQRD constructs a shadow process that reshapes the action space without compromising solution quality. Empirical results demonstrate that in small-scale scenarios, the method reduces candidate actions by 75.9% while preserving 90.8% coverage of near-optimal solutions; in large-scale settings, it achieves the highest intent satisfaction rate and utility, along with the lowest hard constraint violation rate and latency, accompanied by an average 73.0% reduction in action candidates.

Abstract

Modern network policy control maps intent to sequential placement-control decisions. Bellman-style policy optimization primarily asks which action to optimize, while constraints are commonly handled through penalty, barrier, or Lagrangian mechanisms. We observe that before a value function can certify the best deployment, intermediate signals may already identify many candidates that should be excluded from further optimization. This motivates a complementary direction: \emph{Learning Not to Optimize}. Before a value function is accurate enough to select the best placement-control decision, intermediate signals may already show that candidates are equivalent under state--intent relabeling (quotienting), lead to a uniformly worse future state (dominance), or violate executable network laws (residual screening). \LNOQRD{} uses these computed or learned signals as a shadow process to reshape the domain on which primal policy optimization is performed, thereby reducing the action space. We prove lossless quotienting and dominance under explicit equivariance and monotonicity conditions, bound frontier size and ranking cost, and quantify losses from approximate certificates and primal estimates. Experiments show that \LNOQRD{} reduces small-instance candidates by $75.9\%$ while retaining $90.8\%$ near-oracle coverage and, on large instances, achieves the highest utility and intent satisfaction, the lowest hard-law violation and post-generation latency, and a $73.0\%$ average reduction among candidate-based baselines.

Problem

• Intent-Based Network Control
• Action-Space Reduction
• Policy Optimization
• Physics-Informed Constraints
• Non-Optimization Learning

Innovation

• action-space reshaping
• physics-informed constraints
• intent-based control
• non-optimization learning
• quotienting and dominance