FAIRLEARNING
Algorithms AIMech:FairLearning DRAFT
Aliases: Kusner counterfactual fair learning
A causal-model-dependent learning procedure that minimizes predictive loss averaged over inferred background variables, restricting the latent predictor to those variables and observed non-descendants of protected attributes.
Identity Notes & Concept Scope
The named training algorithm is distinct from CounterfactualFairness, the causal model supplied to it and any particular fitted predictor. The source distinguishes the latent predictor from its posterior-averaged reported prediction; this record preserves that distinction. 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
Given a causal model and each training case, infer the posterior of background variables conditioned on observed features and protected attributes. Approximate the expected loss with MCMC samples when needed, expand the training cases with these latent samples, and fit a predictor using the sampled background variables and observed non-descendants. The paper reports predictions by posterior-averaging the fitted latent predictor for a new case.
⚠️ Curated Limitations & Negative Results
- A causal model is an input, and compatible observed distributions need not identify its counterfactual assumptions.
- Lemma 1 applies to functions of non-descendants in the posited model. This record does not turn that lemma into an unconditional guarantee for an arbitrary deployed posterior-mean predictor or a misspecified model.
- Posterior inference and Monte Carlo or optimization approximations introduce computational and estimation dependencies; no exact finite-sample guarantee is asserted.
- Protected attributes and descendant features can inform latent posterior inference even though the latent predictive function is restricted; excluding a raw feature alone is a different procedure.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:CausalGraphicalModel
Causal Graphical Model
|
related_match | Requires a causal graphical model to infer background variables and identify non-descendants. It is a learning algorithm using the model, not a causal model family or an AIO subclass claim. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
original-paper
|
Matt J. Kusner, Joshua R. Loftus, Chris Russell, Ricardo Silva
|
primary paper
full_text
|
ID: arXiv:1703.06856
sha256:2d7371209e44…
|
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/FairLearning.yaml