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

Diffusion models showcased strong capabilities in image synthesis, being usedin many computer vision tasks with great success. To this end, we propose toexplore a new use case, namely to copy black-box classification models withouthaving access to the original training data, the architecture, and the weightsof the model, \ie~the model is only exposed through an inference API. Morespecifically, we can only observe the (soft or hard) labels for some imagesamples passed as input to the model. Furthermore, we consider an additionalconstraint limiting the number of model calls, mostly focusing our research onfew-call model stealing. In order to solve the model extraction task given theapplied restrictions, we propose the following framework. As training data, wecreate a synthetic data set (called proxy data set) by leveraging the abilityof diffusion models to generate realistic and diverse images. Given a maximumnumber of allowed API calls, we pass the respective number of samples throughthe black-box model to collect labels. Finally, we distill the knowledge of theblack-box teacher (attacked model) into a student model (copy of the attackedmodel), harnessing both labeled and unlabeled data generated by the diffusionmodel. We employ a novel active self-paced learning framework to make the mostof the proxy data during distillation. Our empirical results on two data setsconfirm the superiority of our framework over two state-of-the-art methods inthe few-call model extraction scenario.

Towards Few-Call Model Stealing via Active Self-Paced Knowledge Distillation and Diffusion-Based Image Generation