Current Approximate Graph Pattern Mining (AGPM) systems rely on uniform and static sampler allocation strategies, which result in significant computational redundancy because most regions of the sampling space contribute little to the final estimate. However, existing systems overlook this critical characteristic and fail to account for the hierarchical memory architecture of modern servers, leading to slow convergence and substantial memory access overhead. To address these challenges, we propose HierMine, a novel AGPM system designed to overcome the aforementioned drawbacks. The core contributions of HierMine include: (i) a hierarchical sampling strategy that minimizes redundant computation and accelerates convergence; (ii) a dynamic sampling adjustment mechanism that partitions the sampling space online and reallocates samplers adaptively to enhance the hierarchical strategy; (iii) an online grouping-based convergence detection technique that enables the fine-grained dynamic sampling adjustment mechanism; and (iv) a hierarchical data layout optimized for memory access efficiency. Extensive experiments show that HierMine delivers an average speedup of up to 28.9× over state-of-the-art AGPM systems such as ScaleGPM, while maintaining strong theoretical guarantees on estimation quality. It also consistently outperforms exact graph pattern mining systems, highlighting the practical benefits of our approximate approach.

Existing infrared small target detection (IRSTD) methods mainly rely on the assumption that the training and testing data come from the same distribution, a premise that does not hold in many real-world scenarios. Additionally, the inability to access source domain data in numerous IRSTD tasks further complicates the domain adaptation process. To address these challenges, we propose a novel Source-Free Domain Adaptation (SFDA) framework for IRSTD, denoted as IRSTD-SFDA. This framework comprises two key components: Multi-expert Domain Adaptation (MDA) and Multi-scale Focused Learning (MFL). MDA leverages the source model to generate pseudo masks for the target domain, facilitating the transfer of knowledge from the source to the target domain. To account for the inherent diversity of small targets across domains, MDA refines these pseudo masks through a series of operations, including target localization, rolling guidance filtering, shape adaptation, and multi-expert decision, thereby mitigating morphological discrepancies between the source and target domains. Meanwhile, MFL employs a global-local fusion strategy to focus on critical regions, enhancing the model's ability to detect small infrared targets. Extensive experimental evaluations across various cross-domain scenarios demonstrate the effectiveness of the proposed framework.