Singular Value Decomposition (SVD), Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Autoencoders, Fourier and Wavelet Transforms.
Card 1 of 259. Answer: Singular Value Decomposition (SVD), Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Autoencoders, Fourier and Wavelet Transforms.
About · Quizzes · History of AI · To suggest new questions or report an error send me a dm.
Three standard answers are principal component analysis (PCA), singular value decomposition (SVD) and linear discriminant analysis (LDA); autoencoders, t-SNE and UMAP are also common. Dimensionality reduction maps many features to fewer while keeping the useful structure, which cuts noise, speeds up training and makes data plottable.
PCA projects data onto the orthogonal directions of maximum variance. It is usually computed from the SVD of the centered data, X = UΣVᵀ, keeping the first k columns of V. LDA is supervised: it finds directions that best separate known classes, at most C − 1 of them for C classes. Autoencoders learn a non-linear compression through a narrow hidden layer. Fourier and wavelet transforms re-express a signal so a few coefficients carry most of it.
Interviewers often ask how to choose k for PCA: keep enough components to explain, say, 95% of the variance, after standardizing features so the largest-scale one does not dominate. Also distinguish feature extraction, which builds new features, from feature selection, which keeps a subset of the originals.
