English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Given a set of unlabeled images or (image, text) pairs, contrastive learningaims to pre-train an image encoder that can be used as a feature extractor formany downstream tasks. In this work, we propose EncoderMI, the first membershipinference method against image encoders pre-trained by contrastive learning. Inparticular, given an input and a black-box access to an image encoder,EncoderMI aims to infer whether the input is in the training dataset of theimage encoder. EncoderMI can be used 1) by a data owner to audit whether its(public) data was used to pre-train an image encoder without its authorizationor 2) by an attacker to compromise privacy of the training data when it isprivate/sensitive. Our EncoderMI exploits the overfitting of the image encodertowards its training data. In particular, an overfitted image encoder is morelikely to output more (or less) similar feature vectors for two augmentedversions of an input in (or not in) its training dataset. We evaluate EncoderMIon image encoders pre-trained on multiple datasets by ourselves as well as theContrastive Language-Image Pre-training (CLIP) image encoder, which ispre-trained on 400 million (image, text) pairs collected from the Internet andreleased by OpenAI. Our results show that EncoderMI can achieve high accuracy,precision, and recall. We also explore a countermeasure against EncoderMI viapreventing overfitting through early stopping. Our results show that itachieves trade-offs between accuracy of EncoderMI and utility of the imageencoder, i.e., it can reduce the accuracy of EncoderMI, but it also incursclassification accuracy loss of the downstream classifiers built based on theimage encoder.

EncoderMI: Membership Inference against Pre-trained Encoders in Contrastive Learning