Ding Weiping, Lin Chin-Teng, Cao Zehong
IEEE Trans Cybern. 2019 Jul;49(7):2744-2757. doi: 10.1109/TCYB.2018.2834390. Epub 2018 May 22.
Attribute reduction with many patterns and indicators has been regarded as an important approach for large-scale data mining and machine learning tasks. However, it is extremely difficult for researchers to inadequately extract knowledge and insights from multiple overlapping and interdependent fuzzy datasets from the current changing and interconnected big data sources. This paper proposes a deep neuro-cognitive co-evolution for fuzzy attribute reduction (DNCFAR) that contains a combination of quantum leaping particle swarm optimization with nearest-neighbor memeplexes. A key element of DNCFAR resides in its deep neuro-cognitive cooperative co-evolution structure, which is explicitly permitted to identify interdependent variables and adaptively decompose them in the same neuro-subpopulation, with minimizing the complexity and nonseparability of interdependent variables among different fuzzy attribute subsets. Next DNCFAR formalizes to the different types of quantum leaping particles with nearest-neighbor memeplexes to share their respective solutions and deeply cooperate to evolve the assigned fuzzy attribute subsets. The experimental results demonstrate that DNCFAR can achieve competitive performance in terms of average computational efficiency and classification accuracy while reinforcing noise tolerance. Furthermore, it can be well applied to clearly identify different longitudinal surfaces of infant cerebrum regions, which indicates its great potential for brain disorder prediction based on fMRI.
具有多种模式和指标的属性约简已被视为大规模数据挖掘和机器学习任务的一种重要方法。然而，对于研究人员来说，从当前不断变化且相互关联的大数据源中那些多重重叠且相互依存的模糊数据集中充分提取知识和见解极其困难。本文提出了一种用于模糊属性约简的深度神经认知协同进化算法（DNCFAR），它结合了量子跳跃粒子群优化算法和最近邻记忆复合体。DNCFAR的一个关键要素在于其深度神经认知协作协同进化结构，该结构被明确允许识别相互依存的变量，并在同一个神经子群体中对它们进行自适应分解，同时最小化不同模糊属性子集之间相互依存变量的复杂性和不可分离性。接下来，DNCFAR将具有最近邻记忆复合体的不同类型量子跳跃粒子形式化，以共享各自的解决方案并深度协作来进化指定的模糊属性子集。实验结果表明，DNCFAR在平均计算效率和分类准确率方面能够实现具有竞争力的性能，同时增强了噪声容忍能力。此外，它可以很好地应用于清晰识别婴儿大脑区域的不同纵向表面，这表明其在基于功能磁共振成像的脑部疾病预测方面具有巨大潜力。
