MultiHypothesis

Multihypothesis refers to the practice of maintaining and evaluating multiple competing explanations, models, or data associations. It is used across statistics, signal processing, and artificial intelligence to handle uncertainty, ambiguity, or clutter. In statistics, multihypothesis testing concerns evaluating several null hypotheses simultaneously. Controlling error rates is essential, with methods to preserve the familywise error rate (FWER) such as Bonferroni and Holm procedures, and to control the false discovery rate (FDR) such as Benjamini–Hochberg. The approach acknowledges that multiple tests increase the chance of spuriously significant results, and corrective procedures adjust significance thresholds accordingly.

In signal processing and surveillance, multi-hypothesis tracking (MHT) is an algorithmic framework for maintaining a set of competing data associations between observations and tracked objects. It builds a hypothesis tree representing possible tracks, evaluates each hypothesis by a likelihood score, and prunes unlikely branches to manage complexity. MHT is robust in cluttered environments and in situations where measurements are ambiguous or missing, and it has found use in radar, computer vision, and autonomous systems.

Beyond statistics and tracking, the multihypothesis approach underpins reasoning under uncertainty in AI, Bayesian model averaging, and ensemble methods where multiple models or explanations are considered jointly rather than committing to a single hypothesis prematurely.

Practical challenges include combinatorial explosion of hypotheses, computational cost, and the need for effective pruning, gating, and approximation. Success depends on accurate probabilistic models, good priors, and scalable algorithms.

See also: multiple testing, false discovery rate, data association, Bayesian model averaging.