New paper maps 22 fairness notions into near-complete hierarchy for AI resource allocation
Automated inference engine reveals which fairness guarantees hold across goods, chores, and mixed manna.
Deep Dive
Jugal Garg and Eklavya Sharma's new paper organizes 22 fairness notions into a near-complete implication hierarchy, proving or disproving implications for nearly every pair. It covers allocation of goods, chores, and mixed manna under additive, submodular, and subadditive valuations. The work includes an inference engine available as a user-friendly web application that automates part of the analysis and may have broader applications beyond fair division.
Key Points
- 22 fairness notions systematically mapped with proofs or counterexamples for nearly every pair.
- Covers goods, chores, and mixed manna under additive, submodular, and subadditive valuation classes.
- Includes an inference engine web app that automates implication proofs, extendable beyond fair division.
Why It Matters
Provides a rigorous hierarchy so AI systems can guarantee fairness properties without redundant verification across allocation settings.