Home / Curated / Algorithms / Fairness LP

Fairness LP

Algorithms AIMech:FairnessLP DRAFT

Aliases: Dwork fairness linear program

A finite-population optimization procedure that assigns outcome distributions to individuals to minimize expected loss under pairwise Lipschitz similarity constraints, expressible as a linear program for the total-variation or relative-infinity distribution distances used by Dwork et al.

Identity Notes & Concept Scope The source names this optimization formulation Fairness LP. It is a procedure enforcing LipschitzIndividualFairness, not the criterion itself, a generic linear-program solver or a learned similarity metric. The two documented distribution-distance choices are retained as configurations of one formulation. All 442 pinned AIO labels/scoped synonyms and 99 preceding native identities were compared, together with hidden/ignored-inclusive searches of both checkouts; source citations and pending frontier mentions were distinguished from existing concept identities.

Claims & Grounded Evidence

1 Claim

Mechanism #method-scope

Choose a probability vector for each individual and minimize the supplied expected loss. Enforce normalization, nonnegativity and distribution distances bounded by the supplied task metric for every individual pair. Total-variation constraints admit a linear-program formulation; the relative-infinity choice yields pairwise linear probability-ratio bounds with fixed metric-dependent coefficients.

supports Source: Fairness Through Awareness · Locator: Section 2 and Section 2.1, Figure 1 equations (2)–(4), Lemmas 2.1–2.2 and Remark 2.1, PDF pp. 5–6 (printed pp. 4–5)

⚠️ Curated Limitations & Negative Results

The individual metric and utility loss are inputs; solving the program does not establish that either is normatively appropriate.
The cited polynomial-size result concerns a finite set of individuals and outcomes with the specified distribution distances, not arbitrary metrics on output distributions or an unrestricted population.
Extrapolating from sampled individuals to unseen individuals is a separate challenge explicitly noted in Remark 2.1.
Individual constraints do not automatically imply demographic parity without additional assumptions about the population and metric.

Artificial Intelligence Ontology (AIO) Grounding

Mapped AIO Classes

AIO Target Class	Mapping Relation	Mapping Rationale
`AIO:Classification` Classification	related_match	Computes a randomized classification assignment. Classification denotes the task, while this record denotes its constrained optimization procedure; MetricLearning instead learns representations and does not denote this supplied-metric program.

Evidence Sources & Primary Literature

1 Source

Source ID	Publication / Document	Type / Access	Reference & Checksum
`original-paper`	Fairness Through Awareness Cynthia Dwork, Moritz Hardt, Toniann Pitassi, Omer Reingold, Richard Zemel	primary paper full_text	ID: arXiv:1104.3913 sha256:0a5afde6e5a2…

Curation Provenance

Curator: Codex

Method: ai_assisted

Model: GPT-6

Created: 2026-10-11T04:57:48Z

Updated: 2026-10-11T04:57:48Z

Source File: data/curated/FairnessLP.yaml