New algorithm ensures perpetual fairness without knowing future
Fairness slack grows only as sqrt(t), even for unbounded horizons...
A new paper by Kahana, Segal-Halevi, and Hazon tackles a fundamental challenge in online fairness: how to ensure fair outcomes when decisions are irreversible, requests arrive sequentially, and the total duration is unknown. The authors introduce a general framework based on deficits—measuring each fairness requirement's current shortfall relative to a time-varying benchmark. Their simple rule picks, at each round, the action that best improves the next round's deficit profile.
The key theoretical result is an anytime guarantee: after every round, all tracked requirements are satisfied up to a slack that grows only on the order of sqrt(t) (up to logarithmic factors). They prove this growth is unavoidable in general. The framework is instantiated for online allocation of indivisible goods (yielding natural relaxations of proportionality and envy-freeness) and for online public decision-making. Crucially, the algorithm requires no knowledge of the horizon, maximum item value, or any future information—making it practical for applications like food banks, computing resource scheduling, and repeated public decisions.
- Algorithm uses deficit-based metrics to track fairness against time-varying benchmarks
- Slack grows as sqrt(t) (up to log factors), proven to be optimal in general
- Applied to online indivisible goods allocation and public decision-making without horizon knowledge
Why It Matters
Enables perpetual fairness guarantees for long-running automated systems making real-time decisions without future foresight.