Memory-Guided Tree Search with Cross-Branch Knowledge Transfer for LLM Solver Synthesis
The paper introduces MEMOIR, a memory-guided tree-search framework designed for combinatorial optimization problems. It enhances solver synthesis by allowing cross-branch knowledge transfer, improving solution validity and consistency. The framework demonstrates significant performance improvements across various optimization tasks compared to existing methods.
- ▪MEMOIR achieves 96.7% solution validity, outperforming the strongest baseline by 9.2 points.
- ▪The framework includes a two-level memory hierarchy for effective knowledge transfer.
- ▪MEMOIR shows a run-to-run validity standard deviation significantly lower than all evaluated baselines.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.17539 (cs) [Submitted on 17 May 2026] Title:Memory-Guided Tree Search with Cross-Branch Knowledge Transfer for LLM Solver Synthesis Authors:Fatemeh Haji, Javier Delarosa Quiros, Peyman Najafirad View a PDF of the paper titled Memory-Guided Tree Search with Cross-Branch Knowledge Transfer for LLM Solver Synthesis, by Fatemeh Haji and 2 other authors View PDF HTML (experimental) Abstract:Combinatorial optimization (CO) underlies decision-making from logistics to chip design, where infeasible solutions are operationally unusable and small quality gains translate into substantial economic value.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
