MENTAL MODEL · M8186

Counterfactual Regret Minimization

DecidemediumDecision Science

Included

Updated 2026-08-11

Loading revision record…

INTRODUCTION

English translation pending.

CORE DEFINITION

Counterfactual regret minimization is a family of algorithms for solving sequential games: at each information set it computes the regret, how much better the player would have done by choosing the alternative action, and updates the strategy in proportion to that accumulated regret. Over many iterations the average strategy converges to a Nash equilibrium, which is why it became the standard approach for imperfect-information games where direct equilibrium computation is intractable.

SCAFFOLDING EFFECT

psychology

Reduce cognitive load

- Track the regret: record at each decision point how much the missed move would have gained. - Weight the strategy: play actions in proportion to their accumulated regret. - Iterate to equilibrium: average the strategies rather than trusting any single round.

anchor

Anchor fast decisions

Regret is a gradient in strategy space, pointing from what was done toward what should have been done. Updating in proportion to it moves the strategy along that gradient, and because each player's update conditions on the other's current play, the joint dynamics converge to a profile where no player's regret points anywhere useful, which is the definition of equilibrium.

MINIMUM ACTION

In progress 0/1

Practice this model in one real situation:

Track counterfactual regret at each information set of the sequential game. 2. Update the strategy weighted by accumulated regret. 3. Iterate until the strategies converge. 4. Use self-play to generate the iterations without needing an opponent.

Check to track your progress (stored locally)

Learning progress0%

account_treeGenealogyexpand_more

menu_bookReferencesexpand_more

Source support: Explicit

link
doi.orghttps://doi.org/10.5555/1838206.1838229ZH · Explicit
verified

RELATED MODELS