Enabling Federated Inference via Unsupervised Consensus Embedding
Enabling Federated Inference via Unsupervised Consensus Embedding
Authors: Yui Hashimoto, Takayuki Nishio, Yuichi Kitagawa, Takahito Tanimura Date: 2026-05-07 Paper ID: openalex:2605.05718
Summary
The paper introduces Consensus Embedding-based Federated Inference (CE-FI), a framework for distributed model cooperation that preserves privacy by avoiding the exchange of raw data or model parameters. CE-FI uses a learned consensus embedding layer to map heterogeneous model outputs into a shared space, trained exclusively on unlabeled data. Empirical results on image classification benchmarks show that CE-FI maintains strong predictive performance under non-IID conditions while relaxing common assumptions required by existing federated inference methods.
Key Contributions
- Introduces CE-FI, a framework enabling cooperative inference across heterogeneous models without shared parameters, raw data, or encoders.
- Employs a Consensus Embedding (CE) layer to map diverse intermediate representations into a unified space via shared unlabeled data.
- Demonstrates consistent performance gains over solo inference across CIFAR-10 and CIFAR-100 benchmarks under non-IID conditions.
Open Questions & Future Work
- optimizing-ensemble-strategies-for-federated-inference-in-heterogeneous-settings
- privacy-risks-of-shared-intermediate-representations
Key Concepts
- consensus-embedding-based-federated-inference-ce-fi: A framework for federated inference that aligns heterogeneous model representations into a common embedding space using unlabeled data.
Archivist Review
The paper proposes a novel, architecture-agnostic approach to federated inference. I have approved the framework itself as a foundational concept for distributed cooperation and two critical open questions concerning the performance-privacy trade-offs inherent in sharing intermediate latent representations. Standard datasets were rejected as they are routine in the literature.
Approved Concepts
- Consensus Embedding-based Federated Inference (CE-FI): This is the core contribution of the paper, enabling cooperative inference across heterogeneous models while preserving privacy.
Approved Open Questions
- Optimizing Ensemble Strategies in Federated Inference: The ensemble strategy is a primary determinant of performance in federated inference settings where individual models exhibit non-IID characteristics.
- Privacy Risks of Shared Representations: Quantifying the privacy leakage risk of intermediate representations is fundamental to the deployment of federated inference in sensitive domains.
Links
Metadata & Links
- url
- https://arxiv.org/abs/2605.05718
- paper_id
- 2605.05718
- paper_source
- openalex
- domain
- nlp
- tags
-
- architectures
-
- datasets
-
- concept_slugs
- consensus-embedding-based-federated-inference-ce-fi
- dataset_slugs
-
- skill
- TimeSeriesSkill
- created_at
- 2026-05-10T06:05:48Z