Weng and Lin 2011 - Bayesian Approximation for Online Ranking

Routing Summary

Full capture of Weng & Lin (JMLR 2011): closed-form online Bayesian skill updates under Bradley-Terry, Thurstone-Mosteller and Plackett-Luce, the Glicko derivation, and Halo 2 results. 7 notes.

Concept Map

ConceptNoteTypeDepends OnKey Result
W-Stein identityWeng and Lin 2011 - Ranking Models and the W-Stein ApproximationtheoremOverviewposterior moments from
updateWeng and Lin 2011 - Update Rules Derivation and Error CorrectiondefinitionW-Stein,
BT full/partialWeng and Lin 2011 - Bradley-Terry Update AlgorithmstheoremElo with uncertain step size
TM / PLWeng and Lin 2011 - Thurstone-Mosteller and Plackett-Luce Update RulestheoremBTTM k=2 = TrueSkill
GlickoWeng and Lin 2011 - Glicko Derivation and Comparison with ElotheoremBT5-step approximation
ExperimentsWeng and Lin 2011 - Halo 2 Experiments and ResultsexampleallBT-full ~ TrueSkill, 10x faster

Notes

Sources

Weng and Lin 2011 - A Bayesian Approximation Method for Online Ranking

See Also

The Kalman Filter · Reward Modeling from Human Preferences