A Surrogate-Assisted U-Net Architecture Evolutionary Approach Based on GNN

视网膜血管切片在诊断和治疗疾病状况方面起着重要作用。基于识别神经网络的技术在这一领域取得了许多不同的突破，但如何设计最佳的网络结构是一个具有挑战性的问题。神经结构搜索在现实世界的程序中面临着挑战，例如太多的搜索空间和太长的学习时间。为了有效地解决这些问题，该数据显示了基于先进算法的U-Net架构搜索方法。与添加网络参数不同，这种方法可以提高网络性能。同时，该架构结合了代理模型和进化算法，以降低搜索过程中的计算成本。本文使用神经网络作为替代模型来增强变异算法，并通过绘制共同对象来执行匹配措施。同时，本文加速了遗传计算方法的收敛速度。实验结果表明，算法在灵敏度和AUROC等关键指标上仍优于最先进的方法。Retinal vascular sections play an important role in diagnosing and treatment of disease conditions. The technology based on recognition neural networks has made many breakthroughs in this field, but designing the optimal network structure is a challenging problem. Neural structure search faces challenges in real-world programs, such as too much search space and too long learning time. To effectively address these issues, this data demonstrates a U-Net architecture search method based on advanced algorithms. Unlike adding network parameters, this method can improve network performance. Meanwhile, this architecture combines surrogate models and evolutionary algorithms to reduce computational costs during the search process. This article uses neural networks as an alternative model to enhance mutation algorithms and performs matching measures by drawing common objects. Meanwhile, this article accelerates the convergence speed of genetic computing methods. The experimental results show that the algorithm still outperforms state-of-the-art methods in key indicators such as sensitivity and AUROC.