Pith. sign in

REVIEW 2 cited by

Towards Model Agnostic Federated Learning Using Knowledge Distillation

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2110.15210 v2 pith:ZQU5CRVV submitted 2021-10-28 cs.LG cs.DCmath.OCstat.ML

Towards Model Agnostic Federated Learning Using Knowledge Distillation

classification cs.LG cs.DCmath.OCstat.ML
keywords federatedlearningdatadistillationheterogeneityknowledgemodelprotocols
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Is it possible to design an universal API for federated learning using which an ad-hoc group of data-holders (agents) collaborate with each other and perform federated learning? Such an API would necessarily need to be model-agnostic i.e. make no assumption about the model architecture being used by the agents, and also cannot rely on having representative public data at hand. Knowledge distillation (KD) is the obvious tool of choice to design such protocols. However, surprisingly, we show that most natural KD-based federated learning protocols have poor performance. To investigate this, we propose a new theoretical framework, Federated Kernel ridge regression, which can capture both model heterogeneity as well as data heterogeneity. Our analysis shows that the degradation is largely due to a fundamental limitation of knowledge distillation under data heterogeneity. We further validate our framework by analyzing and designing new protocols based on KD. Their performance on real world experiments using neural networks, though still unsatisfactory, closely matches our theoretical predictions.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. IR3DE: A Linear Router for Large Language Models

    cs.CL 2026-06 unverdicted novelty 5.0

    IR3DE is a ridge regression router for domain-expert LLMs that matches or exceeds baselines in language modeling and reasoning tasks while allowing dynamic expert addition or removal without retraining.

  2. Knowledge Distillation in Federated Learning: a Survey on Long Lasting Challenges and New Solutions

    cs.LG 2024-06 unverdicted novelty 2.0

    A survey organizing knowledge distillation techniques for addressing privacy, heterogeneity, communication, and personalization challenges in federated learning.