Neural Cluster First, Route Second: One-Shot Capacitated Vehicle Routing via Differentiable Optimal Transport
Researchers from the academic community have introduced Neural CFRS, a novel framework for solving the Capacitated Vehicle Routing Problem (CVRP), which is critical for modern last-mile logistics. Published on arXiv, this study challenges current Neural Combinatorial Optimization methods that rely on autoregressive decoding, often suffering from sequential bottlenecks and poor out-of-distribution performance. Instead, the authors revisit the classical Cluster-First-Route-Second paradigm, aligning it with deep learning strengths in global context assignment. The proposed model utilizes a differentiable entropic Optimal Transport layer to enforce fleet-capacity constraints end-to-end, producing a continuous transport plan. This non-autoregressive, one-shot approach offers theoretical guarantees against spatial and permutation symmetries. Equipped with a pre-trained spatial vocabulary, Neural CFRS demonstrates extreme parameter efficiency and zero-shot scaling capabilities. It achieves robust performance on large-scale instances with 1000 nodes, maintaining less than a 4% optimality gap, and delivers a competitive 2.73% gap on standard size-100 benchmarks. This advancement highlights significant potential for improving efficiency in logistics operations through lightweight, scalable neural architectures.
Wire timeline
Neural Cluster First, Route Second: One-Shot Capacitated Vehicle Routing via Differentiable Optimal Transport
Researchers from the academic community have introduced Neural CFRS, a novel framework for solving the Capacitated Vehicle Routing Problem (CVRP), which is critical for modern last-mile logistics. Published on arXiv, this study challenges current Neural Combinatorial Optimization methods that rely on autoregressive decoding, often suffering from sequential bottlenecks and poor out-of-distribution performance. Instead, the authors revisit the classical Cluster-First-Route-Second paradigm, aligning it with deep learning strengths in global context assignment. The proposed model utilizes a differentiable entropic Optimal Transport layer to enforce fleet-capacity constraints end-to-end, producing a continuous transport plan. This non-autoregressive, one-shot approach offers theoretical guarantees against spatial and permutation symmetries. Equipped with a pre-trained spatial vocabulary, Neural CFRS demonstrates extreme parameter efficiency and zero-shot scaling capabilities. It achieves robust performance on large-scale instances with 1000 nodes, maintaining less than a 4% optimality gap, and delivers a competitive 2.73% gap on standard size-100 benchmarks. This advancement highlights significant potential for improving efficiency in logistics operations through lightweight, scalable neural architectures.
cs.AI updates on arXiv.org