# Beyond Aggregate Drift: Optimal Client-Specific Windows in Federated Learning

arXiv: https://arxiv.org/abs/2610.15543 · HTML: https://mlsanity.com/paper/2610.15543/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 11 of its 21 references that are in the map.
- First posted on arXiv 11 Oct 2026.

## Builds on (most cited first, 11 of 11)

- [Online Learning under Delayed Feedback](https://mlsanity.com/paper/1306.0686/) (2013)
- [Federated Learning under Distributed Concept Drift](https://mlsanity.com/paper/2206.00799/) (2022)
- [Federated Learning for Data Streams](https://mlsanity.com/paper/2301.01542/) (2023)
- [An Adaptive Algorithm for Learning with Unknown Distribution Drift](https://mlsanity.com/paper/2305.02252/) (2023)
- [Stochastic Optimization under Distributional Drift](https://mlsanity.com/paper/2108.07356/) (2021)
- [Online Federated Learning via Non-Stationary Detection and Adaptation amidst Concept Drift](https://mlsanity.com/paper/2211.12578/) (2022)
- [Tight Time Complexities in Parallel Stochastic Optimization with Arbitrary Computation Dynamics](https://mlsanity.com/paper/2408.04929/) (2024)
- [Nonparametric Density Estimation under Distribution Drift](https://mlsanity.com/paper/2302.02460/) (2023)
- [LASG: Lazily Aggregated Stochastic Gradients for Communication-Efficient Distributed Learning](https://mlsanity.com/paper/2002.11360/) (2020)
- [Streaming Federated Learning with Markovian Data](https://mlsanity.com/paper/2503.18807/) (2025)
- [Adaptive Data Admission and Retention for Streaming Federated Learning](https://mlsanity.com/paper/2607.23987/) (2026)

## Abstract (arXiv)

We study federated learning from continuously arriving data: each client samples from its own distribution, these distributions drift over time, and the server must track the minimizer of the clients' weighted average loss. Clients are not alike as they collect data at different rates, with different noise, and change at different speeds. The main question is how much history should each one keep? A single shared window must serve every client at once, and so makes a fast-changing client carry the sampling noise of a slow, nearly stationary one. We give each client a memory of its own, sized to that client's noise, arrival rate, and drift, and refresh it by restarted accelerated method that starts from a common point. The tracking error is then bounded by [ Rleft(summ wm3/2,σm√γm/λmright)2/3, ] for clients with weight wm, noise σm, arrival rate λm, and drift rate γm on a domain of diameter R. This is never worse than what any common window can achieve, and strictly better whenever noise and drift are spread unevenly across clients. A lower bound that puts different groups of clients on different drift clocks shows this quantity is the right one. The two bounds agree up to a constant whenever the clients occupy boundedly many refresh scales, and in general differ only by the sixth root of the number of scales. Experiments show the gain comes from the memory rather than from the solver: it persists across accelerated, stochastic-gradient, and one-step methods, and against a separately tuned common-memory baseline.