基于萤火虫算法的脑效应连接网络学习方法

Learning effective connectivity network structure based on firefly algorithm

摘要:

脑效应连接网络学习是人脑连接组研究的一个重要研究课题,准确识别脑效应连接网络对于脑疾病的早期诊断以及病理研究具有重要意义.本文将萤火虫算法与贝叶斯网相结合,提出了一种带有繁殖机制的脑效应连接网络萤火虫学习方法.新方法使用K2评分作为目标函数来衡量萤火虫个体的绝对亮度,利用萤火虫种群的寻优来完成脑效应连接网络的学习,并利用繁殖机制对种群实施进一步的优化.首先将一种仅含少数边的脑效应连接网络表示成一个萤火虫个体,并通过萤火虫个体的定向移动操作以及随机移动操作逐步构建脑效应连接网络；然后每经过一定代数的寻优后,萤火虫种群执行一次繁殖过程,以优化效应连接网络的质量.最后,当算法收敛时,将萤火虫种群中绝对亮度最高个体所代表的网络结构作为学习到的最优脑效应连接网络.在多组模拟数据集上的实验结果验证了新算法中繁殖机制的有效性,且与其它算法相比,新算法具有明显优势.在真实数据上的实验也表明了算法的潜在实用性.

Abstract:

Learning brain effective connectivity (EC) networks is an important topic within the community of human brain connectome. It is of great significance for the early diagnosis and pathological study of brain diseases to accurately identify the brain EC network structure. This paper combines the Firefly Algorithm (FA) with Bayesian network, and proposes a new method to learn brain EC networks by FA with a reproductive mechanism. The new method uses K2 score as the evaluation method of absolute brightness of fireflies, uses the optimization of firefly population to complete the learning of brain EC networks, and uses reproductive mechanism to further optimize the population. First, a firefly individual represented a brain EC network with a few edges, which was gradually constructed through the directional movements and random movements of the firefly individual. Then, a reproductive mechanism was employed to optimize the quality of networks after a certain number of evolution iterations. Finally, the network structure represented by the individuals with the highest absolute brightness in the population was used as the learning brain EC network. Experimental results on many simulated datasets verified the effectiveness of the reproductive mechanism, and the new algorithm has obvious advantages on the whole performance compared with other algorithms. Experimental results on real datasets also show the potential practicability of the new algorithm.