Detecting Atypical Clients in Federated Learning via Representation-Level Divergence
A new research paper proposes a lightweight geometric signal to detect atypical clients in federated learning by measuring how local training alters the activation-induced partition of the input space on a shared probe set. This method yields a permutation-invariant, interpretable metric of client-global divergence that captures differences in data processing, addressing challenges from heterogeneous data, distributional shifts, and anomalous inputs.
Key facts
- arXiv:2605.22266v1
- Announce Type: cross
- Proposes a lightweight geometric signal
- Quantifies functional deviation of a client with respect to the global model
- Measures how local training alters activation-induced partition of input space
- Evaluated on a shared probe set
- Yields permutation-invariant, interpretable metric
- Captures differences in how data is processed by the model
Entities
—