Dice Question Streamline Icon: https://streamlinehq.com

Sample-complexity of learning bounded contracts under weaker regularity

Investigate whether similar polynomial sample-complexity guarantees for learning bounded contracts can be obtained under weaker regularity assumptions than combined FOSD and CDFP, for example under only FOSD or only CDFP.

Information Square Streamline Icon: https://streamlinehq.com

Background

The survey demonstrates polynomial sample-complexity (and corresponding online regret) guarantees for learning bounded contracts when both FOSD and CDFP hold. It explicitly points out uncertainty about relaxing these assumptions.

Progress here would broaden the applicability of learning-based contract design to more general technologies and outcome models.

References

It remains an open question whether analogous sample complexity results can be obtained under weaker regularity assumptions (e.g., only one of FOSD or CDFP).

— Algorithmic Contract Theory: A Survey (2412.16384 - Duetting et al., 20 Dec 2024) in Section 6.3 (Improved Regret Bounds under Regularity Assumptions)