SCAFFLSA: Quantifying and Eliminating Heterogeneity Bias in Federated Linear Stochastic Approximation and Temporal Difference Learning
Mangold, Paul, Samsonov, Sergey, Labbi, Safwan, Levin, Ilya, Alami, Reda, Naumov, Alexey, Moulines, Eric
–arXiv.org Artificial Intelligence
In this paper, we perform a non-asymptotic analysis of the federated linear stochastic approximation (FedLSA) algorithm. We explicitly quantify the bias introduced by local training with heterogeneous agents, and investigate the sample complexity of the algorithm. We show that the communication complexity of FedLSA scales polynomially with the desired precision $\epsilon$, which limits the benefits of federation. To overcome this, we propose SCAFFLSA, a novel variant of FedLSA, that uses control variates to correct the bias of local training, and prove its convergence without assumptions on statistical heterogeneity. We apply the proposed methodology to federated temporal difference learning with linear function approximation, and analyze the corresponding complexity improvements.
arXiv.org Artificial Intelligence
Feb-6-2024
- Country:
- Asia > Middle East
- UAE (0.14)
- Europe (0.28)
- Asia > Middle East
- Genre:
- Research Report (0.64)
- Industry:
- Education (0.54)
- Technology: