MetaFed: Meta-Level Coordination in FL

• MetaFed is a federated learning paradigm that employs decentralized meta-level coordination to manage resource orchestration in heterogeneous environments.
• It integrates reinforcement learning, homomorphic encryption, and carbon-aware scheduling to optimize model accuracy, privacy, and environmental impact.
• In healthcare settings, MetaFed enables server-free cyclic knowledge distillation among federations, effectively addressing non-i.i.d. data challenges and personalization.

Searching arXiv for the named MetaFed papers and related usage. The term MetaFed has been used on arXiv to denote more than one federated-learning framework rather than a single canonical method. In one usage, MetaFed is a fully decentralized federated-learning framework tailored for Metaverse environments, designed to jointly address performance, privacy, and environmental sustainability through intelligent multi-agent orchestration, privacy-preserving aggregation, and carbon-aware scheduling (Yagiz et al., 24 Aug 2025). In another usage, MetaFed is a server-free framework for federated learning among federations, where each federation is treated as a meta-distribution and knowledge is transferred cyclically to produce personalized models under feature shift and label shift (Chen et al., 2022). A related line of work, "FedMeNF," explicitly situates its contribution within a broader MetaFed paradigm for federated meta-learning, especially in privacy-sensitive neural-field settings (Yun et al., 8 Aug 2025). Taken together, these works use the name to mark decentralized or meta-level coordination mechanisms in federated systems, but they target different problem formulations, architectures, and optimization objectives.

1. Terminological scope and problem settings

In the 2025 Metaverse paper, MetaFed is introduced to address the limits of centralized architectures in immersive environments, where latency sensitivity, rich multimodal data, energy consumption, and privacy constraints interact directly. The stated objective is to support sustainable and intelligent resource orchestration in decentralized federated learning, with simultaneous attention to accuracy, privacy guarantees, and carbon footprint (Yagiz et al., 24 Aug 2025).

In the 2022 healthcare-oriented paper, MetaFed addresses a different problem: federated learning among different federations when federations may distrust one another or lack a central server. The framework is defined around the goal of learning a personalized model $f_i:\mathcal{X}\to\mathcal{Y}$ for each federation $F_i$, under non-i.i.d. data distributions $P_i\neq P_j$, while avoiding raw-data sharing and centralized coordination (Chen et al., 2022).

These two uses share a decentralization motif but differ in the unit of coordination. In the Metaverse setting, coordination occurs among heterogeneous edge and cloud resource providers and clients; in the healthcare setting, coordination occurs among entire federations. This suggests that "MetaFed" functions less as a single algorithmic family than as a recurring label for decentralized, meta-level federated optimization schemes.

2. MetaFed for federated Metaverse systems

The Metaverse formulation defines MetaFed as a system built around three pillars: intelligent multi-agent orchestration, privacy-preserving model updates, and carbon-aware scheduling (Yagiz et al., 24 Aug 2025). Its infrastructure consists of three components.

First, Resource Providers $(\mathcal{R})$ form a dynamic pool of heterogeneous edge and cloud nodes. Each provider $r_i$ is represented as

$r_i=\langle C_i, N_i, E_i, L_i\rangle,$

where $C_i$ is normalized compute capability, $N_i$ is network bandwidth, $E_i$ is an energy-efficiency score, and $L_i$ is geo-location for emission modeling.

Second, Orchestration Agents $F_i$0 are decentralized reinforcement-learning agents, one per provider, that determine which clients to activate in each communication round and how to weight them.

Third, the Federated Registry $F_i$1 is a lightweight, blockchain-backed distributed hash table that stores resource metadata and cryptographic identities, enabling tamper-proof discovery without a central broker.

The round structure is explicit. Agents observe local and global signals, choose a slate of clients, and instruct them to train locally. Clients then encrypt and upload masked model updates. An edge or cloud aggregator computes an encrypted sum and publishes an updated global model. The paper therefore places client selection, aggregation security, and energy-aware scheduling inside a single orchestration loop rather than treating them as separate systems concerns (Yagiz et al., 24 Aug 2025).

3. Optimization mechanisms in the Metaverse formulation

The client-selection mechanism is based on tabular Q-learning with $F_i$2-greedy exploration. At time $F_i$3, the joint system state is

$F_i$4

where $F_i$5 is the current carbon-intensity class, $F_i$6 encodes the recent trend in global model accuracy, and $F_i$7 aggregates convergence velocity and utilization history. The policy is

$F_i$8

with geometrically decaying exploration,

$F_i$9

The reward combines model quality, efficiency, and emissions: $P_i\neq P_j$0 with $P_i\neq P_j$1, $P_i\neq P_j$2, and $P_i\neq P_j$3. To favor greener nodes, the Q-value update includes an environmental correction,

$P_i\neq P_j$4

where $P_i\neq P_j$5 and $P_i\neq P_j$6.

Privacy preservation is implemented through additive homomorphic encryption, exemplified by Paillier, together with differential-privacy noise. Each client computes a local model update $P_i\neq P_j$7 and encrypts it as

$P_i\neq P_j$8

The aggregator computes

$P_i\neq P_j$9

after which a coordinating server or trusted coordinator decrypts the result: $(\mathcal{R})$0 Global model updating follows a weighted FedAvg rule,

$(\mathcal{R})$1

To address statistical heterogeneity, the framework allows a client-side FedProx regularizer,

$(\mathcal{R})$2

with $(\mathcal{R})$3 (Yagiz et al., 24 Aug 2025).

Carbon-aware scheduling is modeled through a sinusoidal grid carbon-intensity process,

$(\mathcal{R})$4

with $(\mathcal{R})$5, $(\mathcal{R})$6, and $(\mathcal{R})$7. The provider priority is

$(\mathcal{R})$8

where $(\mathcal{R})$9. By construction, client selection is discounted when carbon intensity is high, aligning the FL process with cleaner energy availability.

4. Empirical profile of the Metaverse formulation

The experimental setup uses MNIST and CIFAR-10, a lightweight "ResNet Tiny" (RT) with 4.8 million parameters, and a federated protocol with 50 simulated clients, Dirichlet partitioning with $r_i$0, 20% client participation per round, 5 local epochs, batch size 32, and 100 communication rounds (Yagiz et al., 24 Aug 2025). The reported metrics are test accuracy, per-round and cumulative CO$r_i$1 emissions, average communication time per round, and total bytes transmitted.

On MNIST, the full configuration MetaFed (RL + Green + RT) reaches 99.60% accuracy, 337.6 gCO$r_i$2/round, 45 846 g cumulative emissions, and 33.9 s/round. Against FedAvg at 99.19% accuracy and 578.4 gCO$r_i$3/round, this corresponds to a 41.6% per-round emission reduction and a 0.41% accuracy gain (Yagiz et al., 24 Aug 2025).

On CIFAR-10, MetaFed(RL + Green + RT) achieves 80.26% accuracy, 287.9 gCO$r_i$4/round, 45 634 g total emissions, and 30.3 s/round. Compared to FedAvg at 66.56% accuracy and 575.7 gCO$r_i$5/round, the reported difference is a 20 pp accuracy improvement and a 50% per-round emissions cut.

The paper’s broader summary states that MetaFed delivers up to 25% lower CO$r_i$6 emissions compared with conventional FL across the training lifecycle, while also maintaining high accuracy and modest communication overhead. The coexistence of the "up to 25%" lifecycle statement with larger per-round reductions on MNIST and CIFAR-10 indicates that the paper distinguishes between aggregate lifecycle accounting and benchmark-specific per-round comparisons. A plausible implication is that the stronger per-round reductions reflect particular dataset and scheduling conditions, whereas the 25% figure summarizes the overall training lifecycle under the paper’s evaluation protocol.

5. Server-free MetaFed among federations

The 2022 MetaFed paper formulates a different federated-learning problem in which there are $r_i$7 federations $r_i$8, each with a private dataset

$r_i$9

partitioned into training, validation, and test splits (Chen et al., 2022). The assumptions are non-i.i.d. distributions, shared input and output spaces, and no central server. Each model factorizes as

$r_i=\langle C_i, N_i, E_i, L_i\rangle,$0

where $r_i=\langle C_i, N_i, E_i, L_i\rangle,$1 is a feature extractor and $r_i=\langle C_i, N_i, E_i, L_i\rangle,$2 is a classifier head.

The defining mechanism is Cyclic Knowledge Distillation. Each federation is treated as a meta-distribution $r_i=\langle C_i, N_i, E_i, L_i\rangle,$3, and model parameters are passed cyclically,

$r_i=\langle C_i, N_i, E_i, L_i\rangle,$4

At federation $r_i=\langle C_i, N_i, E_i, L_i\rangle,$5, the teacher is $r_i=\langle C_i, N_i, E_i, L_i\rangle,$6, and training uses a feature-distillation loss

$r_i=\langle C_i, N_i, E_i, L_i\rangle,$7

An optional posterior-based cyclic KD loss is also given: $r_i=\langle C_i, N_i, E_i, L_i\rangle,$8 where $r_i=\langle C_i, N_i, E_i, L_i\rangle,$9. The federation-level training loss is

$C_i$0

Training alternates between common knowledge accumulation and personalization. In the first stage, $C_i$1 cyclic rounds are run across federations. At federation $C_i$2, the incoming teacher is validated on $C_i$3. If its accuracy exceeds $C_i$4, the local model is initialized randomly and trained with fixed $C_i$5; otherwise it is initialized from the teacher and trained with the same loss. After $C_i$6 rounds, the last model $C_i$7 is treated as the common model and broadcast to all federations.

In the personalization stage, each federation compares the common model’s validation performance $C_i$8 against its own local validation score $C_i$9. If $N_i$0 or $N_i$1, then $N_i$2 and the method reduces to pure fine-tuning. Otherwise, the knowledge-distillation weight is set dynamically as

$N_i$3

This design makes personalization contingent on whether the common model is helpful for a given federation, rather than enforcing a uniform cross-federation transfer policy (Chen et al., 2022).

6. Communication, empirical results, and relation to broader MetaFed ideas

For the server-free healthcare formulation, each cyclic pass communicates one model per federation, yielding $N_i$4 model transfers per round, and the total number of transfers is $N_i$5 when including the personalization pass. Local computation per transfer is given as $N_i$6, and overall complexity is $N_i$7, which the paper describes as being on par with FedAvg’s $N_i$8 (Chen et al., 2022). Empirically, the method is reported to require fewer rounds $N_i$9 for convergence under non-i.i.d. data than FedAvg, with communication reduced by up to 50%.

The reported results cover both feature-shift and label-shift regimes. On VLCS, MetaFed achieves 64.27 average accuracy versus FedBN 60.02. On PACS, it achieves 64.06 versus 59.83. On PAMAP2 cross-person, it reaches 86.07 versus 85.03. Under label shift, the paper reports PAMAP2 Dirichlet split at 90.07 versus 87.44, MedMNIST results of 96.15/92.07/91.28 versus 92.32/89.07/78.37, Real COVID-19 X-ray at 91.99 versus FedProx 87.31, and Parkinson’s tremor at 87.42 versus FedBN 80.94. The paper states that all methods use identical architectures, SGD with learning rate $E_i$0, and shared split percentages, and that MetaFed consistently outperforms FedAvg, FedProx, and FedBN (Chen et al., 2022).

Ablation results attribute part of the gain to both training stages: removing the common-knowledge stage or removing personalization degrades accuracy by 3–5%, and replacing KD with fine-tuning yields a similar drop. The paper also reports that under limited total rounds $E_i$1, MetaFed retains $E_i$2 accuracy while FedAvg and FedBN drop to $E_i$3. A variant called MetaFed++, based on grouping 20 federations into 3 groups and running intra-group then inter-group MetaFed, improves average accuracy by 1–2% over flat MetaFed (Chen et al., 2022).

A related but distinct development is "FedMeNF: Privacy-Preserving Federated Meta-Learning for Neural Fields" (Yun et al., 8 Aug 2025). Although not itself titled MetaFed, its detailed description explicitly discusses a broader MetaFed paradigm. FedMeNF frames federated meta-learning around a meta-parameter $E_i$4, client-level support and query sets, and a privacy-preserving outer-loop loss,

$E_i$5

introduced to regulate privacy leakage in local meta-optimization. Its theoretical discussion states that privacy leakage, measured through $E_i$6, is bounded by $E_i$7 under the first-order approximation used in the paper, while retaining a gradient-alignment term for fast adaptation. This suggests a further extension of the MetaFed label: from server-free or decentralized orchestration toward privacy-aware federated meta-learning for highly personalized representations such as neural fields.

Across these works, the name MetaFed consistently denotes federated systems that move beyond standard server-centric FedAvg. In one line, it refers to decentralized orchestration with reinforcement learning, homomorphic encryption, and carbon-aware scheduling for Metaverse infrastructure (Yagiz et al., 24 Aug 2025). In another, it refers to cyclic, server-free knowledge aggregation and personalization among federations (Chen et al., 2022). In the broader context evoked by FedMeNF, it also points toward federated meta-learning with explicit privacy-utility control (Yun et al., 8 Aug 2025). The unifying theme is decentralized or meta-level coordination under heterogeneity, but the concrete algorithms, assumptions, and evaluation criteria are domain-specific rather than interchangeable.