TCT: Convexifying Federated Learning using Bootstrapped Neural Tangent Kernels
This addresses optimization challenges in federated learning for neural networks, particularly in non-IID settings, offering a method to improve model performance for distributed clients.
The paper tackles the performance gap between federated and centralized learning when clients have dissimilar data distributions, showing that nonconvexity causes final layers to fail in utilizing learned features. It proposes a Train-Convexify-Train (TCT) procedure using neural tangent kernels, achieving accuracy improvements of up to +36% on FMNIST and +37% on CIFAR10.
State-of-the-art federated learning methods can perform far worse than their centralized counterparts when clients have dissimilar data distributions. For neural networks, even when centralized SGD easily finds a solution that is simultaneously performant for all clients, current federated optimization methods fail to converge to a comparable solution. We show that this performance disparity can largely be attributed to optimization challenges presented by nonconvexity. Specifically, we find that the early layers of the network do learn useful features, but the final layers fail to make use of them. That is, federated optimization applied to this non-convex problem distorts the learning of the final layers. Leveraging this observation, we propose a Train-Convexify-Train (TCT) procedure to sidestep this issue: first, learn features using off-the-shelf methods (e.g., FedAvg); then, optimize a convexified problem obtained from the network's empirical neural tangent kernel approximation. Our technique yields accuracy improvements of up to +36% on FMNIST and +37% on CIFAR10 when clients have dissimilar data.