Training privacy-preserving video analytics pipelines by suppressing features that reveal information about private attributes

TL;DR

This paper proposes a training method for neural networks that reduces private attribute leakage during inference by using a confusion loss, effectively balancing privacy preservation with task accuracy.

Contribution

It introduces a novel training approach that suppresses private attribute information in features, enhancing privacy in video analytics without significantly harming performance.

Findings

Reduces gender attribute leakage by 2.88%

Decreases age group leakage by 13.06%

Maintains high task accuracy with minimal impact

Abstract

Deep neural networks are increasingly deployed for scene analytics, including to evaluate the attention and reaction of people exposed to out-of-home advertisements. However, the features extracted by a deep neural network that was trained to predict a specific, consensual attribute (e.g. emotion) may also encode and thus reveal information about private, protected attributes (e.g. age or gender). In this work, we focus on such leakage of private information at inference time. We consider an adversary with access to the features extracted by the layers of a deployed neural network and use these features to predict private attributes. To prevent the success of such an attack, we modify the training of the network using a confusion loss that encourages the extraction of features that make it difficult for the adversary to accurately predict private attributes. We validate this training approach on image-based tasks using a publicly available dataset. Results show that, compared to the original network, the proposed PrivateNet can reduce the leakage of private information of a state-of-the-art emotion recognition classifier by 2.88% for gender and by 13.06% for age group, with a minimal effect on task accuracy.