Non-Negative Matrix Factorization (NMF) for Dimensionality Reduction in Image Data

<p>I have already discussed different types of dimensionality reduction techniques in detail.</p> <p><strong>Principal Component Analysis (PCA)</strong>,&nbsp;<strong>Factor Analysis (FA)</strong>,&nbsp;<strong>Linear Discriminant Analysis (LDA)</strong>,&nbsp;<strong>Autoencoders (AEs)</strong>, and&nbsp;<strong>Kernel PCA</strong>&nbsp;are the most popular ones.</p> <p>Non-Negative Matrix Factorization (NMF or NNMF) is also a&nbsp;<em>linear</em>&nbsp;dimensionality reduction technique that can be used to reduce the dimensionality of the feature matrix.</p> <p>All dimensionality reduction techniques fall under the category of unsupervised machine learning in which we can reveal hidden patterns and important relationships in the data without requiring labels.</p> <p>So, dimensionality reduction algorithms deal with unlabeled data. When training such an algorithm, the&nbsp;<strong>fit()</strong>&nbsp;method only needs the feature matrix,&nbsp;<strong>X</strong>&nbsp;as the input and it does not require the label column,&nbsp;<strong>y</strong>.</p> <p>As its name implies, non-negative matrix factorization (NMF) needs the feature matrix to be non-negative.</p> <p>Because of this non-negativity constraint, the usage of NMF is limited to data with non-negative values such as image data (pixel values always lie between 0 and 255, hence there are no negative values in image data!).</p> <p><a href="https://towardsdatascience.com/non-negative-matrix-factorization-nmf-for-dimensionality-reduction-in-image-data-8450f4cae8fa">Read More</a></p>