FedReason: Large Language Models as Weight-Space Meta-Aggregators for Heterogeneous Federated Learning
Opens the thread: an LLM as a training-free meta-aggregator in the FL server.
3 items · 2 sources · 2 days
Operational story trace
Follow in this browser to see new updates on your Live feed.
Latest change
Two Nov 12 papers move agents into FL operations: Cortex-FL uses agents to find collaborators and coordinate training, and a companion paper proposes scientific checks to catch FL code that silently drifts from its protocol.
This is not one event: it is three separate arXiv papers that apply LLMs or agents to federated learning (FL). The first, FedReason on Nov 5, asks whether an LLM can inspect client models and pick the server's aggregation rule without any training.
Arc
Opens the thread: an LLM as a training-free meta-aggregator in the FL server.
Adds the failure mode: FL implementations can diverge from protocol while results still look plausible.
Extends agents from the server to the full collaboration workflow, including partner discovery and incentives.
Opens the thread: an LLM as a training-free meta-aggregator in the FL server.
Adds the failure mode: FL implementations can diverge from protocol while results still look plausible.
Extends agents from the server to the full collaboration workflow, including partner discovery and incentives.
What to watch — open questions
Storylines are threaded mechanically from the feed: stories that share a distinctive anchor across multiple days and sources. Each item links to its original source. The evidence trace, current state, and open questions are written by the editor routine and refreshed whenever a new beat lands.