Deep learning model construction for a semi-supervised classification with feature learning

Abstract Several deep models were proposed in image processing, data interpretation, speech recognition, and video analysis. Most of these architectures need a massive proportion training samples use arbitrary configuration. This paper constructs learning architecture with feature learning. Graph convolution networks (GCNs), semi-supervised graph representation, have become increasingly popular as cost-effective efficient methods. existing merging node descriptions for distribution on the stabilised neighbourhood knowledge, typically requiring significant amount variables high degree computational complexity. To address concerns, this research presents DLM-SSC, unique method classification tasks that can combine knowledge from multiple neighbourhoods at same time by integrating high-order employs two function techniques reducing number parameters hidden layers: modified marginal fisher analysis (MMFA) kernel principal component (KPCA). The MMFA KPCA weight matrices are layer when implementing DLM, supervised pretraining technique doesn't require lot information. Free measuring citation datasets (Citeseer, Pubmed, Cora) other sets demonstrate suggested approaches outperform similar algorithms.