keyword

semi-nonnegative matrix factorization

Semi-nonnegative matrix factorization is a dimensionality reduction and data representation technique that approximates a matrix as the product of two lower-rank factor matrices, requiring only one of the factors to contain strictly non-negative values while allowing the input data and the second factor to have mixed signs. Unlike standard non-negative matrix factorization, which demands that all data and resulting matrices be non-negative, this relaxation allows the technique to be applied directly to centered, normalized, or real-valued datasets containing negative numbers. In practice, the unconstrained factor functions as a set of basis vectors that capture structural patterns across positive and negative values, while the non-negative factor acts as an interpretable weight or soft-clustering assignment matrix that describes how the original data points are composed.

1 item