← AI Terminology
Majority Voting
Majority voting aggregates multiple model outputs by choosing the most frequent discrete answer (or label) among them.
It is the aggregation rule inside self-consistency and many ensemble systems.
It is the aggregation rule inside self-consistency and many ensemble systems.
Why It Matters in AI
When answers are discrete (A/B/C, numbers, yes/no), voting is a robust, training-free ensemble. It underpins self-consistency and multi-agent debate tie-breaks.
Key Points
| Aspect | Description |
|---|---|
| Need | Comparable, parseable final answers |
| Rule | Mode of answers; optional confidence weighting |
| Weakness | Ties; correlated errors; free-form text hard to vote |
| Classic ML | Same idea as ensemble voting classifiers |
| Extensions | Weighted vote by logprob or verifier score |
| Vs RM pick | No learned scorer required |
Simple Analogy
A show of hands in a room: the option with the most hands wins — simple democracy over model samples.
Common Usage Examples
- Self-consistency majority over boxed answers
- Ensemble classifiers in classical ML
- Multi-agent systems: vote on a plan
- Break ties with highest log-prob sample
Summary
In short: Majority voting picks the most common answer among many samples — simple ensemble aggregation for discrete outputs.