CADO: From Imitation to Cost Minimization for Heatmap-based Solvers in Combinatorial Optimization

2026-02-09

AI Summary

This work addresses the objective mismatch in existing heatmap-based combinatorial optimization solvers, where minimizing imitation loss during supervised training does not guarantee cost-optimal solutions. To bridge this gap, the authors formulate the diffusion denoising process as a Markov decision process and introduce a Label-Centered Reward mechanism coupled with a Hybrid Fine-Tuning strategy. By leveraging reinforcement learning to directly optimize the cost of decoded solutions, the approach effectively mitigates both Decoder-Blindness and Cost-Blindness, achieving end-to-end cost-aware objective alignment. Experimental results across multiple combinatorial optimization benchmarks demonstrate that the proposed method attains state-of-the-art performance, underscoring the critical role of objective alignment in enhancing the efficacy of heatmap-based solvers.

Technology Category

• Search and Optimization: Learning to Search
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization
• Computer Vision: Learning & Optimization for CV

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Heatmap-based solvers have emerged as a promising paradigm for Combinatorial Optimization (CO). However, we argue that the dominant Supervised Learning (SL) training paradigm suffers from a fundamental objective mismatch: minimizing imitation loss (e.g., cross-entropy) does not guarantee solution cost minimization. We dissect this mismatch into two deficiencies: Decoder-Blindness (being oblivious to the non-differentiable decoding process) and Cost-Blindness (prioritizing structural imitation over solution quality). We empirically demonstrate that these intrinsic flaws impose a hard performance ceiling. To overcome this limitation, we propose CADO (Cost-Aware Diffusion models for Optimization), a streamlined Reinforcement Learning fine-tuning framework that formulates the diffusion denoising process as an MDP to directly optimize the post-decoded solution cost. We introduce Label-Centered Reward, which repurposes ground-truth labels as unbiased baselines rather than imitation targets, and Hybrid Fine-Tuning for parameter-efficient adaptation. CADO achieves state-of-the-art performance across diverse benchmarks, validating that objective alignment is essential for unlocking the full potential of heatmap-based solvers.

Problem

• Combinatorial Optimization
• Heatmap-based Solvers
• Objective Mismatch
• Cost Minimization
• Supervised Learning

Innovation

• Cost-Aware Diffusion
• Objective Alignment
• Reinforcement Learning Fine-tuning
• Heatmap-based Solvers
• Label-Centered Reward