MultiFair: Multi-Criteria Fairness Framework

• MultiFair is a suite of frameworks that simultaneously enforce multiple fairness criteria in diverse settings such as resource allocation and fairness-aware classification.
• It integrates axiomatic concepts like envy-freeness, proportionality, EF1/EFX, and maximin share with efficient algorithms to tackle computational and optimization challenges.
• Its applications span multi-agent allocation, federated learning, multimodal diagnosis, and market-oriented systems, demonstrating broad practical and theoretical impact.

Searching arXiv for the cited papers to ground the article. MultiFair denotes a family of fairness frameworks in which several fairness dimensions must be satisfied simultaneously rather than one at a time. In the fair-allocation literature, it is best understood as fairness for many agents sharing goods, chores, or resources, viewed through multiple fairness criteria at once, with emphasis on existence, efficiency, computational complexity, and algorithmic techniques across indivisible goods, divisible resources, chores, mixed items, matching, and public decision-making (Aziz, 2019). In later work, closely related uses of the term extend to fairness-aware classification, federated learning, multimodal medical diagnosis, online recommendation, auctions, bandits, and hierarchical allocation, while retaining the common idea of jointly balancing multiple protected attributes, groups, tasks, agents, objectives, or structural constraints (Swati et al., 9 Sep 2025, Zubair et al., 30 Sep 2025, Lucet et al., 30 Dec 2025).

1. Scope and uses of the term

Within multi-agent allocation, MultiFair refers to a toolkit that combines axiomatic fairness notions such as envy-freeness, proportionality, EF1/EFX, and maximin share with efficiency notions such as Pareto efficiency and Nash social welfare, together with market-style mechanisms such as CEEI and DRF (Aziz, 2019). Outside allocation, the label is used more broadly. MMM-fair defines multi-fairness as the simultaneous consideration of multiple protected attributes, multiple fairness definitions, and multiple optimization objectives within one training and model-selection workflow (Swati et al., 9 Sep 2025). In multimodal medical classification, MultiFair names a framework that jointly addresses modality learning bias and demographic fairness bias by dual-level gradient modulation (Zubair et al., 30 Sep 2025). In hierarchical allocation, multilevel fair allocation studies tree-structured hierarchical relations among agents and asks how local mechanisms can retain fairness and efficiency throughout the tree (Lucet et al., 30 Dec 2025).

Domain	Representative work	Meaning of “MultiFair”
Multi-agent allocation	(Aziz, 2019)	fairness for many agents sharing goods, chores, or resources, viewed through multiple fairness criteria at once
Fairness-aware classification	(Swati et al., 9 Sep 2025)	simultaneous consideration of multiple protected attributes, multiple fairness definitions, and multiple optimization objectives
Multi-label classification	(Liu et al., 2023)	group-fair decision rules across a vector of correlated targets
Federated learning	(Su et al., 2023)	simultaneous group fairness and client fairness
Multimodal medical classification	(Zubair et al., 30 Sep 2025)	joint balancing of modality contributions and demographic fairness
Hierarchical allocation	(Lucet et al., 30 Dec 2025)	fairness and efficiency across tree-structured allocations

Taken together, these usages suggest that “MultiFair” functions less as a single standardized formalism than as an umbrella label for settings in which fairness is intrinsically multi-criteria.

2. Axiomatic foundations in multi-agent fair allocation

The canonical allocation model considers a set of agents $N=\{1,\dots,n\}$ and a set $G$ of items or resources. Items may be indivisible or divisible. For goods, valuations are typically normalized and monotone, and a central class is additive valuations, where for indivisible goods

$v_i(S)=\sum_{g\in S} v_i(\{g\}).$

For chores, disutilities $d_i$ are often used, or one works with nonpositive valuations; many results extend to mixed manna settings containing both goods and chores (Aziz, 2019).

The core comparison-based notion is envy-freeness. For an allocation $A=(A_1,\dots,A_n)$,

$$\text{EF:}\quad \forall i,j\in N,\quad v_i(A_i)\ge v_i(A_j).$$

Proportionality requires each agent to receive at least a $1/n$ share of the value of the whole resource,

$$\text{PROP:}\quad \forall i\in N,\quad v_i(A_i)\ge \frac{1}{n}v_i(G).$$

Because exact EF may fail for indivisible goods, two relaxations dominate the literature: $$\text{EF1:}\quad \forall i,j\in N,\ \exists g\in A_j\ \text{s.t.}\ v_i(A_i)\ge v_i(A_j\setminus\{g\}),$$ and

$$\text{EFX:}\quad \forall i,j\in N,\ \forall g\in A_j\ \text{with}\ v_i(\{g\})>0,\quad v_i(A_i)\ge v_i(A_j\setminus\{g\}).$$

EFX is strictly stronger than EF1, and its general existence for indivisible goods remains largely open beyond special cases (Aziz, 2019).

The principal share-based notion is the maximin share. For agent $G$0,

$G$1

and an allocation is MMS-fair if $G$2 for all $G$3. Exact MMS allocations may fail to exist even under additive valuations, which is why much of the literature studies approximation (Aziz, 2019).

Efficiency enters through Pareto efficiency and welfare maximization. Pareto efficiency requires that no feasible allocation weakly improve every agent and strictly improve at least one. Nash social welfare maximizes

$G$4

For positive additive goods, NSW-maximizing allocations are EF1 and Pareto efficient, making NSW a central fairness-efficiency compromise. In divisible settings, CEEI yields envy-free and Pareto efficient allocations in many classical cases, while leximin remains a canonical equity objective (Aziz, 2019).

3. Allocation settings, algorithms, and recent extensions

The survey literature presents a sharply differentiated landscape across domains. For indivisible goods, EF often fails, checking whether an EF allocation exists is NP-complete even for $G$5–$G$6 utilities, and maximizing egalitarian welfare or NSW is NP-hard. Under additive valuations, however, simple sequential round-robin algorithms achieve EF1, and with positive additive utilities NSW maximization yields allocations that are both EF1 and Pareto efficient. For divisible goods, envy-free and Pareto optimal allocations exist generally, and for positive additive utilities NSW maximization yields EF and PO in polynomial time. Cake-cutting supports proportional allocations easily, while bounded envy-free protocols are substantially harder; the survey also treats chores, mixed items, rent division, matching, public decision-making, group fairness, and graph-constrained allocations (Aziz, 2019).

More recent work strengthens the simultaneous-fairness agenda. For submodular valuations, there exists a universal constant $G$7 such that every instance admits an allocation that is simultaneously EFL, hence EF1, and $G$8-MMS. The randomized polynomial-time algorithm RECE1 always outputs an EFL allocation, achieves an ex-ante MMS guarantee $G$9, and achieves $v_i(S)=\sum_{g\in S} v_i(\{g\}).$0-MMS for all agents with positive probability ex post (Feige et al., 4 Jun 2026). This extends simultaneous fairness guarantees beyond additive valuations to monotone submodular valuations.

A different structural extension concerns item multiplicities. For a multiset of indivisible items with $v_i(S)=\sum_{g\in S} v_i(\{g\}).$1 distinct types and additive valuations, complete EFX allocations always exist for arbitrary $v_i(S)=\sum_{g\in S} v_i(\{g\}).$2 and $v_i(S)=\sum_{g\in S} v_i(\{g\}).$3 when $v_i(S)=\sum_{g\in S} v_i(\{g\}).$4, with both a constructive polynomial-time proof and a geometric proof. The same work shows that complete EF allocations exist when at least one agent has a unique valuation and the number of items of each type exceeds a particular finite threshold; it also provides explicit upper and lower bounds on that threshold in special cases (Gorantla et al., 2022). These results show that type structure can fundamentally alter existence and tractability.

4. Statistical, multimodal, and federated interpretations

In fairness-aware classification, multi-fairness is explicitly framed as a multi-objective optimization problem. MMM-fair supports parity-style group fairness criteria during training, including demographic parity, equalized odds, and equal opportunity, computes metrics over intersectional groups $v_i(S)=\sum_{g\in S} v_i(\{g\}).$5, and trains boosting ensembles to jointly minimize classification loss and fairness violations: $v_i(S)=\sum_{g\in S} v_i(\{g\}).$6 Its implementation controls fairness strength through $v_i(S)=\sum_{g\in S} v_i(\{g\}).$7, uses a softmax-weighted aggregation of fairness gradients across protected attributes, and exposes model trade-offs through a Pareto frontier explorer (Swati et al., 9 Sep 2025).

In multi-label classification, SimFair extends Demographic Parity and Equalized Opportunity to vector-valued targets by expressing fairness in terms of the predicted probability vector $v_i(S)=\sum_{g\in S} v_i(\{g\}).$8. Because EOp becomes unstable for rare label combinations, SimFair replaces the hard indicator $v_i(S)=\sum_{g\in S} v_i(\{g\}).$9 with a similarity-based weight. Using Jaccard similarity and exponential scaling,

$d_i$0

it defines a similarity-induced fairness constraint that interpolates between DP and EOp. The training objective is

$d_i$1

and the paper shows that DP and EOp are special cases of the framework (Liu et al., 2023).

Federated learning adds another axis: fairness across both groups and clients. mFairFL formulates group fairness through constraints such as DP, EO, and AP, writes the constrained empirical-risk objective as

$d_i$2

and relaxes it with differential multipliers,

$d_i$3

Before aggregation, the method detects conflicts among local gradients and iteratively curates their direction and magnitude, using cosine-similarity targets updated by exponential moving averages (Su et al., 2023).

A broader heterogeneous multi-task generalization appears in FairMT. There, fairness must span classification, detection, and regression under incomplete supervision. The method introduces task-specific asymmetric violations, aggregates them into a unified constraint $d_i$4, and solves the resulting problem by a primal–dual scheme together with a head-aware multi-objective proxy that models head-induced anisotropy (Hu et al., 29 Nov 2025). In multimodal medical classification, MultiFair addresses an adjacent problem: balancing uneven learning across modalities while reducing group disparities. Its dual-level gradient modulation combines modality-level balancing factors,

$d_i$5

with group-level fairness factors derived from group-wise EMA surrogate AUC gaps, and activates fairness modulation only when the fusion model’s group AUC disparity exceeds a threshold $d_i$6 (Zubair et al., 30 Sep 2025).

5. Strategic, online, and market-oriented interpretations

In market design, one strand operationalizes MultiFair through auction objectives that add explicit fairness terms to standard welfare optimization. The multi-dimensional fairness combinatorial double auction model for cloud environments combines quantitative, qualitative, and stochastic fairness. Its winner-determination problem maximizes

$d_i$7

where the fairness factor depends on losses, wins, consecutive losses, bid quality, and stochastic application probabilities. The objective is summarized in the paper as $d_i$8 (Hassanzadeh et al., 2016).

In online recommendation, multi-sided fairness is formulated as a constrained optimization problem. Problem FAIR chooses a recommendation matrix $d_i$9 to maximize expected platform revenue while requiring each item’s realized outcome to be at least a fraction $A=(A_1,\dots,A_n)$0 of its item-fair outcome and each user type’s realized outcome to be at least a fraction $A=(A_1,\dots,A_n)$1 of its user-fair outcome. The resulting interpolation between platform utility, item fairness, and user fairness is paired with FORM, an online algorithm that uses relaxed constraints and persistent exploration; under bandit feedback it achieves

$A=(A_1,\dots,A_n)$2

for revenue regret and fairness regret (Chen et al., 2023).

In distributed learning without communication, “My Fair Bandit” studies $A=(A_1,\dots,A_n)$3 cooperative players and $A=(A_1,\dots,A_n)$4 arms with collisions and unknown utilities. The fairness target is a max-min fair matching

$A=(A_1,\dots,A_n)$5

and the proposed algorithm learns such a matching with regret

$A=(A_1,\dots,A_n)$6

which the paper characterizes as regret optimal up to a $A=(A_1,\dots,A_n)$7 factor (Bistritz et al., 2020). In mixed-motive Markov games, a related but distinct line replaces utilitarian welfare by proportional fairness, defining a fair altruistic utility in log-payoff space and a Fair Markov Game objective

$A=(A_1,\dots,A_n)$8

with actor-critic algorithms built from fair advantages (Xu et al., 9 Feb 2026).

A further generalization appears in multiparty multiobjective optimization. There, fairness concerns not allocation itself but the evaluation of solution sets across heterogeneous decision makers. A concession rate vector $A=(A_1,\dots,A_n)$9 defines a commonly acceptable region, and classical metrics such as IGD or HV are embedded in a Nash-product-based evaluation

$$\text{EF:}\quad \forall i,j\in N,\quad v_i(A_i)\ge v_i(A_j).$$0

which is shown to satisfy Pareto monotonicity, symmetry, balance preference, and acceptability monotonicity (Zhao et al., 30 Jan 2026).

6. Hierarchies, environments, and open directions

A distinct hierarchical interpretation is developed in multilevel fair allocation. The setting is a rooted tree in which the root initially owns all items and allocations are propagated top-down until they reach the leaves. Leaves have matroid-rank utility functions, and each internal node’s utility is the sum of the utilities of its children. The first algorithm, SMA, defines an estimated utility

$$\text{EF:}\quad \forall i,j\in N,\quad v_i(A_i)\ge v_i(A_j).$$1

and then recursively chooses, at each internal node, a local allocation that is both utilitarian-optimal and $$\text{EF:}\quad \forall i,j\in N,\quad v_i(A_i)\ge v_i(A_j).$$2-maximizing with respect to $$\text{EF:}\quad \forall i,j\in N,\quad v_i(A_i)\ge v_i(A_j).$$3. The second algorithm, MGYS, extends General Yankee Swap to the multilevel setting. SMA is polynomial-time and comes with fairness and efficiency guarantees; MGYS preserves multilevel utilitarian optimality and exhibits excellent fairness properties in practice, but not full fairness guarantees (Lucet et al., 30 Dec 2025).

The dynamic-systems literature complements these algorithmic models with testbeds. MAFE introduces Multi-Agent Fair Environments as fairness-aware Dec-POMDPs,

$$\text{EF:}\quad \forall i,j\in N,\quad v_i(A_i)\ge v_i(A_j).$$4

and instantiates the framework in lending, healthcare, and education. These environments use temporal group-fairness metrics, rate-based outcomes, and multiple interacting decision makers such as insurers, hospitals, universities, employers, and planners, thereby making MultiFair a property of system dynamics rather than a single static decision rule (Lazri et al., 25 Feb 2025).

Across these strands, the most persistent open problems remain those already emphasized in the fair-allocation survey: general existence of EFX for indivisible goods, tight approximation bounds and simpler algorithms for MMS for goods and chores, fairness under connectivity, matroid and matching constraints, mixed manna with both goods and chores, and robust mechanisms for dynamic and online environments with limited information (Aziz, 2019). Later domain-specific work adds further frontiers: extending simultaneous guarantees beyond the present valuation classes and tree structures in hierarchical allocation (Lucet et al., 30 Dec 2025), handling intersectional sparsity and broader modalities in fairness-aware learning (Swati et al., 9 Sep 2025, Zubair et al., 30 Sep 2025), and developing principled ways to compare balanced relative gains across multiple parties and objectives (Zhao et al., 30 Jan 2026). In that sense, MultiFair is not one theorem or one mechanism, but a research program centered on the simultaneous management of several fairness claims within a single technical system.