For a decade, “training a model” meant one big machine, or a tightly-coupled cluster, chewing through a clean dataset. That era is ending. Learning is moving to the edge — across phones, vehicles, sensors, and geographically scattered servers that are slow, intermittently connected, resource-starved, and sometimes actively malicious. My group has studied what it takes to learn well under those conditions.
Robustness when some participants lie
In distributed and federated learning, many workers contribute updates and a server aggregates them. What if some workers are compromised and send poisoned updates? Naïve averaging fails badly — a few bad actors can wreck the model. In Byzantine Resilience With Reputation Scores (2022) we defend by having the system learn whom to trust: workers accrue reputation, and the aggregate discounts the untrustworthy. Robustness becomes something the system infers over time, not a fixed assumption.
Learning despite lag
Distributed training is also plagued by asynchrony — workers finish at different times, so updates arrive stale. It’s tempting to think this must hurt generalization. In Distributed SGD Generalizes Well Under Asynchrony (2019) we showed the opposite can hold: done right, asynchronous training still generalizes, which is what makes large-scale distributed learning practical.
Bandits: learning while you act
At the edge you often can’t separate “collect data” from “make decisions” — you must learn while acting, paying for every mistake. That’s the bandit setting, and it shows up everywhere once you look:
- Maximizing success rate of payment routing using non-stationary bandits (2023) — routing each transaction to succeed, in a world whose statistics drift.
- Interference constrained beam alignment for time-varying channels via kernelized bandits (2022) — pointing a wireless beam correctly as the channel changes.
- Weighted Gaussian process bandits for non-stationary environments (2022) — the common thread: learn fast, but forget at the right rate when the world moves.
That last point matters. At the edge, the environment is rarely stationary, so a good learner has to weigh fresh evidence against stale.
Why it matters
The systems we increasingly rely on — recommendation, payments, wireless, federated models trained on our devices — are distributed, adversarial, and non-stationary by nature. Algorithms that assume clean, centralized, i.i.d. data quietly fail in exactly the places they’re deployed. Designing for untrustworthiness, lag, and drift is not hardening an ideal system; it’s building the real one.
What it means going forward
Three forces make edge learning central to the next decade:
- Privacy and data gravity push training toward the data — onto devices — rather than hauling data to a central store.
- Adversaries are now assumed, not hypothetical; poisoning and manipulation are part of the threat model for any open learning system.
- Non-stationarity is the norm as models learn continuously from a world that won’t hold still.
The through-line across all of this — and across much of my group’s theory, from iterated random operators to the convergence guarantees underneath these algorithms — is asking not just does it work? but does it still work when the setting is distributed, adversarial, and drifting? Increasingly, that is the only question that matters.
Papers behind this post: Byzantine Resilience With Reputation Scores (2022) · Distributed SGD Generalizes Well Under Asynchrony (2019) · Maximizing success rate of payment routing using non-stationary bandits (2023) · Interference constrained beam alignment … via kernelized bandits (2022) · Weighted Gaussian process bandits for non-stationary environments (2022). See the Publications page.