The low autocorrelation binary sequences (LABS) problem presents significant challenges in combinatorial optimization, affecting fields such as communications and satellite navigation. A new paper, submitted on June 17, 2026, by authors Blaž Pšeničnik, Borko Bošković, Jan Popić, and Janez Brest, introduces a hybrid search framework aimed at optimizing LABS efficiency.
Innovative Hybrid Search Framework for LABS
The proposed method integrates Thompson sampling with parallel self-avoiding walks. This approach allows for adaptive allocation of computational resources across various restriction classes within the LABS search space. By modeling these partitions as arms in a multi-armed bandit framework, the method dynamically reallocates search resources toward partitions demonstrating higher merit factors, while still exploring less-sampled areas.
Key features of this framework include:
- GPU-parallel execution to accelerate processes
- Shared posterior updates for improved efficiency
- Bloom filter implementation to prevent cycle occurrences
Two-Stage Optimization Strategy
This research employs a two-stage optimization strategy that first examines constrained partitioned skew-symmetric spaces. Following this, the framework refines the best candidates within an unrestricted search area. This two-tiered approach has proven effective in enhancing the performance of LABS optimization.
The experiments conducted on long binary sequences revealed that this method surpasses previously established results for 35 sequence lengths ranging from 450 to 527, and also for L=573. Notably, a new record was established with a merit factor exceeding 8.0 for L=451.
Significance of the Findings
The results highlight the effectiveness of Thompson sampling in prioritizing partitions that yield better observed performance. This reinforces the importance of online, data-driven resource allocation in optimizing LABS. Overall, the framework offers a scalable and efficient approach for maximizing merit factors in binary sequences.
🤖 This article was rewritten by Feed and Figures' editorial AI from a report originally published by arXiv Machine Learning. Facts and quotes are preserved from the original; the rewrite focuses on clarity and structure. For the unedited original, see the source link below.
