Data-Driven Action Recourse via Historical User Trajectories
October 28, 2026
The DDAR framework enables machine learning systems to provide actionable recourse to users without manually designed cost functions. By building dynamics models from historical user data, the method identifies low-cost trajectories toward desired outcomes based on the frequency of observed paths.
HOW THIS AFFECTS YOU
●
researcherYou can bypass expensive human supervision when designing recourse mechanisms for predictive models.