On the Difficulty of Defending Self-Supervised Learning against Model Extraction
This addresses a security threat for ML-as-a-Service providers offering SSL models, but it is incremental as it focuses on adapting known extraction issues to SSL.
The paper tackles the problem of defending self-supervised learning (SSL) models against model extraction attacks, finding that existing defenses are inadequate and attacks using stolen representations are query-efficient and achieve high downstream accuracy.
Self-Supervised Learning (SSL) is an increasingly popular ML paradigm that trains models to transform complex inputs into representations without relying on explicit labels. These representations encode similarity structures that enable efficient learning of multiple downstream tasks. Recently, ML-as-a-Service providers have commenced offering trained SSL models over inference APIs, which transform user inputs into useful representations for a fee. However, the high cost involved to train these models and their exposure over APIs both make black-box extraction a realistic security threat. We thus explore model stealing attacks against SSL. Unlike traditional model extraction on classifiers that output labels, the victim models here output representations; these representations are of significantly higher dimensionality compared to the low-dimensional prediction scores output by classifiers. We construct several novel attacks and find that approaches that train directly on a victim's stolen representations are query efficient and enable high accuracy for downstream models. We then show that existing defenses against model extraction are inadequate and not easily retrofitted to the specificities of SSL.