A Path Integral Surrogate for Multi-Step Gradient Inversion in Federated Learning
PI-SME attacks the hidden multi-step FedAvg path, not just a single gradient point.
The paper proposes treating a client’s accumulated update as a path integral of gradients along a learnable Bézier trajectory. It approximates that integral with Gauss-Legendre quadrature nodes, aiming to better invert updates produced after several local training steps. On CIFAR-100 and FEMNIST, the authors report more faithful private-image reconstructions than the strongest surrogate baseline across several metrics and matching loss. Source: ArXiv · AI/CL/LG's note.
The paper proposes treating a client’s accumulated update as a path integral of gradients along a learnable Bézier trajectory. It approximates that integral with Gauss-Legendre quadrature nodes, aiming to better invert updates produced after several local training steps. On CIFAR-100 and FEMNIST, the authors report more faithful private-image reconstructions than the strongest surrogate baseline across several metrics and matching loss. Source: ArXiv · AI/CL/LG's note.
score 4