Heterogeneous Federated Zeroth-Order Optimization using Gradient Surrogates

Federated optimization, an emerging paradigm that finds wide applications, e.g., federated learning, enables multiple clients (e.g., edge devices) to collaboratively optimize a global function by sharing their local gradients. However, the gradient information is not available in many applications, giving rise to the paradigm of federated zeroth-order optimization (ZOO). Existing federated ZOO algorithms typically suffer from the limitations of query and communication round inefficiency, which can be attributed to (a) their reliance on a substantial number of function queries for gradient estimation and (b) the significant disparity between their realized local updates and the intended global updates caused by client heterogeneity. To this end, we (a) introduce trajectory-informed gradient surrogates which are capable of using the history of function queries during optimization for accurate and query-efficient gradient estimation, and (b) develop the technique of adaptive gradient correction using these surrogates to mitigate the aforementioned disparity. With these, we propose the federated zeroth-order optimization using gradient surrogates (FZooS) algorithm for query- and communication round-efficient heterogeneous federated ZOO, which is supported by our theoretical analyses and extensive experiments.