Discount Yourself: Self-inclusive Freshness-weighted Gossip for Decentralised Learning under Client Outages

Key points

• We observe that the stalest of these models is usually the client's own, a case that existing staleness-aware rules, which discount only the models a client receives, do not address.1
• We formalise aggregation after an outage as estimating the state of a fully available network and derive three principles.1
• We introduce the effective lag, which tracks this quantity at the cost of one scalar per message.1
• These principles define Self-inclusive Freshness-weighted Gossip (SFG), a family of drop-in aggregation rules with the communication and computation of decentralised federated averaging, whose guarantees hold for the stale receiver itself.1

Sources (1)

• [1]Discount Yourself: Self-inclusive Freshness-weighted Gossip for Decentralised Learning under Client Outages
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 05:12 PM

  Decentralised federated learning (DFL) trains personalised models on devices that average with their neighbours rather than with a central server.

  We observe that the stalest of these models is usually the client's own, a case that existing staleness-aware rules, which discount only the models a client receives, do not address.

Extractive summary: sentences quoted from the sources.