Neuro-Evolutionary Heuristics for Variable Gapped Common Subsequence Identification
A recent preprint on arXiv (2608.00888) presents a novel learning-driven technique for the automatic creation of heuristics aimed at addressing the Variable Gapped Longest Common Subsequence Problem (VGLCSP). This problem is a variation of the traditional longest common subsequence issue, incorporating gap constraints and is relevant in fields like sequence alignment and time-series analysis. While dynamic programming effectively handles two-sequence cases, multi-sequence scenarios often depend on beam search heuristics that are manually crafted and lack reliability. The new approach formulates heuristics as a neural network with a set architecture, refined through a genetic algorithm in a neuro-evolutionary context. The iterative process alternates between optimizing weights and conducting evaluations using a cutting-edge multi-source beam search technique. This research seeks to enhance the efficacy and reliability of heuristics by employing data-driven methods. The paper can be accessed on arXiv with the identifier 2608.00888.
Key facts
- The paper addresses the Variable Gapped Longest Common Subsequence Problem (VGLCSP).
- VGLCSP is a variant of the longest common subsequence problem with gap constraints.
- Applications include sequence alignment and time-series analysis.
- Two-sequence versions are solved with dynamic programming.
- Multi-sequence forms are usually solved with beam search-based heuristics.
- The proposed approach uses a neural network with predefined architecture.
- Weights are optimized by a genetic algorithm within a neuro-evolutionary framework.
- The learning process alternates between weight optimization and evaluation in an iterative multi-source beam search.
Entities
Institutions
- arXiv