Reduced cost-based ranking for generating promising subproblems
M. Milano, W. J. van Hoeve
Abstract
In this paper, we propose an effective search procedure that interleaves two steps: subproblem generation and subproblem solution. We mainly focus on the first part. It consists of a variable domain value ranking based on reduced costs. Exploiting the ranking, we generate, in a Limited Discrepancy Search tree, the most promising subproblems first. An interesting result is that reduced costs provide a very precise ranking that allows to almost always find the optimal solution in the first generated subproblem, even if its dimension is significantly smaller than that of the original problem. Concerning the proof of optimality, we exploit a way to increase the lower bound for subproblems at higher discrepancies. We show experimental results on the TSP and its time constrained variant to show the effectiveness of the proposed approach, but the technique could be generalized for other problems.
Create a lesson
Related papers
BEAKER: An Expert-Curated Benchmark for Embodied Brains in Self-Driving Chemical Laboratories
Fei Lin, Tengchao Zhang, Ziyang Gong et al.
FrontierPhysics: Evaluating Agents for End-to-End Physics Research
Bingran You, Wenjun Ke, Ria Rosenauer et al.
Dull, Dirty, Dangerous? Working Conditions and the Limits of Occupational AI Exposure
Siyi Yu
Learning to Decode Linear Codes with Generalized Belief Propagation Network
Samira Malek
On Turn-Level Best-of-N Sampling for Scaling LLM Agents
Jalaj Bhandari, Dhawal Gupta, Hongli Zhan et al.
CompressionBench: How can your coding agents compress LLMs?
Xinrui Zhong, Zhifeng Li, Zhiming Wang et al.