minMM

minMM is an acronym that appears in several disciplines, but there is no universally accepted definition. In optimization and decision theory, minMM is occasionally used informally to denote a “minimum of the maximum” objective, i.e., a minimax problem focused on reducing the worst-case outcome. This is closely related to robust optimization, where the goal is to minimize the worst-case loss.

In statistics and data analysis, the letters MM can stand for “mean mismatch” or “mean difference.” In such contexts, minMM could describe a criterion that selects models, alignments, or parameters by minimizing the mean discrepancy across observations. However, this usage is not standardized and may be specific to a certain paper or software package.

In machine learning and pattern recognition, minMM has been used in notes and code as shorthand for a low-margin or minimum-margin criterion in training classifiers, aiming to maximize resilience to misclassification. Again, this is not a formal, widely adopted term.

Because minMM lacks a single, canonical meaning, interpretation depends on the source. When encountered, it is best to consult the defining document or tool to determine what MM stands for in that context (for example, mean, margin, mismatch, or maximum) and what the objective is.

See also: minimization, minimax, margin in machine learning, mean squared error.