Adapting t-SNE for data streams with S+t-SNE allows for real-time visualization and handling of evolving data dynamics.
This research paper proposes a novel dimension reduction technique called Random Projection Ensemble Dimension Reduction (RPEDR) for high-dimensional regression problems, leveraging an ensemble of carefully selected random projections to efficiently capture the essential information between predictors and response variables.
POLCA Net is a novel autoencoder architecture that effectively performs non-linear dimensionality reduction and feature extraction by combining orthogonality constraints, variance-based feature sorting, and optional class label integration, achieving superior performance in both classification and reconstruction tasks compared to traditional PCA.
POLCA Net 是一種基於自動編碼器的深度學習架構,旨在將 PCA 和 LDA 的優點與非線性映射相結合,以更好地處理複雜數據,並在降維、正交性、基於方差的特徵排序和高保真重建方面表現出色。
This research paper introduces a novel approach called random effects sufficient dimension reduction (SDR) for analyzing clustered data, addressing the limitations of existing SDR methods by accounting for heterogeneity between clusters and incorporating both continuous and time-invariant binary predictors.
This paper introduces a novel online landmark replacement algorithm for out-of-sample dimensionality reduction methods, enhancing the embedding of streaming data by dynamically updating landmarks using geometric graphs and minimal dominating sets.