LINC: Decoupling Local Consequence Scoring from Hidden Matching in Constructive Neural Routing

2026-05-07

AI Summary

This work addresses a key limitation in existing constructive neural routing solvers, which rely solely on matching decoder context with candidate embeddings while ignoring deterministic local consequences such as travel time, waiting time, slack time, and capacity changes. To remedy this, the authors propose the LINC architecture, which explicitly models these local consequences and differentially incorporates them based on their decision-making roles. LINC employs a shared linear local scorer to compare centralized relative consequences and modulates the decoder context using a summary of feasible candidates, thereby preserving the global matching mechanism. This design decouples local scoring from implicit state matching, preventing redundant learning of transition logic. Experiments demonstrate that LINC significantly improves performance, reducing PolyNet’s optimality gaps on CVRPTW from 13.83%/38.15% to 7.26%/14.71% on Solomon/Homberger benchmarks, and achieving state-of-the-art results on TSP and CVRP.

Abstract

Constructive neural routing solvers usually score the next action by matching a decoder context to candidate embeddings, hiding deterministic one-step consequences such as travel, waiting, slack, and capacity changes. We propose LINC (Local Inference via Normed Comparison), a decoder-side candidate decision architecture that computes these consequences explicitly. LINC uses them according to their decision role: centered relative consequences are compared by a shared linear local scorer, while feasible-set summaries modulate the decoder context. This preserves standard global matching and relieves the hidden state from rediscovering transition arithmetic. The Capacitated Vehicle Routing Problem with Time Windows (CVRPTW) serves as the main constrained-routing stress test; the same interface extends to the Capacitated Vehicle Routing Problem (CVRP) and Traveling Salesman Problem (TSP). In particular, for CVRPTW, LINC reduces PolyNet's Solomon/Homberger gaps from 13.83\%/38.15\% to 7.26\%/14.71\%; for TSP and CVRP, it also improves external-benchmark gaps.

Problem

• constructive neural routing
• local consequences
• hidden matching
• Capacitated Vehicle Routing Problem
• Traveling Salesman Problem

Innovation

• constructive neural routing
• local consequence scoring
• decoupling
• normed comparison
• combinatorial optimization