Low-rank decomposition is a linear algebra technique that approximates a matrix as the product of two or more smaller matrices of lower rank, reducing the number of parameters needed to represent it. In machine learning it underlies parameter-efficient fine-tuning methods such as LoRA, which learn a small low-rank update to a pretrained weight matrix instead of updating all parameters. It trades a controlled amount of representational capacity for large reductions in memory and compute cost.