<ul><li>Federated Class-Incremental Learning (FCIL) enables Class-Incremental Learning (CIL) from distributed data.</li><li>Spatial-Temporal Statistics Aggregation (STSA) is proposed to address challenges in FCIL methods, providing a unified framework to aggregate feature statistics spatially and temporally.</li><li>STSA-E is introduced as a communication-efficient variant with theoretical guarantees, achieving similar performance to STSA-E with lower communication overhead.</li><li>Experiments on three FCIL datasets demonstrate that the STSA method outperforms existing methods in terms of performance, flexibility, and communication and computation efficiency.</li></ul>

STSA: Federated Class-Incremental Learning via Spatial-Temporal Statistics Aggregation

Discover more