首页|Study Data from Guangdong University of Technology Update Knowledge of Computati onal Intelligence (A Novel Hypercomplex Graph Convolution Refining Mechanism)
Study Data from Guangdong University of Technology Update Knowledge of Computati onal Intelligence (A Novel Hypercomplex Graph Convolution Refining Mechanism)
扫码查看
点击上方二维码区域,可以放大扫码查看
原文链接
NETL
NSTL
By a News Reporter-Staff News Editor at Robotics & Machine Learning Daily News Daily News-Researchers detail new data in Machine Learning-Computational Intelligence. According to news reporting originating from Guangzhou, People's Republic of China, by NewsRx correspondents, research s tated, "Hypercomplex graph convolutions with higher hypercomplex dimensions can extract more complex features in graphs and features with varying levels of comp lexity are suited for different situation. However, existing hypercomplex graph neural networks have a constraint that they can only carry out hypercomplex grap h convolutions in a predetermined and unchangeable dimension." Financial supporters for this research include National Natural Science Foundati on of China (NSFC), Science and Technology Development Fund of Macau SAR, Guangd ong Basic and Applied Basic Research Foundation, Key Areas Research and Developm ent Program of Guangzhou, Guangdong Provincial Key Laboratory of Cyber-Physical System. Our news editors obtained a quote from the research from the Guangdong University of Technology, "To address this limitation, this paper presents a solution to overcome this limitation by introducing the FFT-based Adaptive Fourier hypercomp lex graph convolution filtering mechanism (FAF mechanism), which can adaptively select hypercomplex graph convolutions with the most appropriate dimensions for different situations by projecting the outputs from all candidate hypercomplex g raph convolutions to the frequency domain and selecting the one with the highest energy via the FFT-based Adaptive Fourier Decomposition. Meanwhile, we apply th e FAF mechanism to our proposed hypercomplex high-order interaction graph neural network (HHG-Net), which performs high-order interaction and strengthens intera ction features through quantum graph hierarchical attention module and feature i nteraction gated graph convolution. During convolution filtering, the FAF mechan ism projects the outputs from different candidate hypercomplex graph convolution s to the frequency domain, extracts their energy, and selects the convolution th at outputs the largest energy. After that, the model with selected hypercomplex graph convolutions is trained again. Our method outperforms many benchmarks, inc luding the model with hypercomplex graph convolutions selected by DARTS, in node classification, graph classification, and text classification."
GuangzhouPeople's Republic of ChinaA siaComputational IntelligenceMachine LearningGuangdong University of Techn ology