You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Learning to Perform Local Rewriting for Combinatorial Optimization [[notes]](./paper_notebook/Learning_to Perform_Local_Rewriting_for_Combinatorial_Optimization.md)
In this work, instead of finding a solution from scratch, we first construct a feasible one, then make incremental improvement by iteratively applying local rewriting rules to the existing solution until convergence.
DRL+CO+生成式:
Paper
Summary
Attention, Learn to Solve Routing Problems! [notes]
对于类似TSP的组合优化问题,通过Encoder生成节点嵌入,然后输入decoder生成策略$\pi$,训练方法是 REINFORCE with a simple baseline based on a deterministic greedy rollout。可以解决TSP、VRP、OP、PCTSP等问题,在多个问题上使用很灵活。
Reinforcement Learning and Additional Rewards for the Traveling Salesman Problem [notes]
对于DL+RL求解TSP的问题,通过在训练过程中传递最小生成树信息极高解的质量。还提出了一种能够实时预测TSP实例的最佳长度的深度学习架构。以Deudon等人提出的TSP机器学习算法+Geometric Deep Learning 为基础,目的是了解如何在能够有效解决其他组合优化问题的丰富框架中混合机器学习方法和组合优化概念。