<ul><li>This paper investigates the risks of inference-time data leakage in deep neural networks (NNs).</li><li>The study focuses on residual NNs, specifically the use of skip connections in residual blocks.</li><li>The authors propose a backward feature inversion method called PEEL to recover input features from intermediate outputs of residual NNs.</li><li>PEEL surpasses state-of-the-art recovery methods in terms of mean squared error (MSE) when tested on facial image datasets and pre-trained classifiers.</li></ul>

PEEL the Layers and Find Yourself: Revisiting Inference-time Data Leakage for Residual Neural Networks

Discover more