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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

This paper investigates the piracy problem of deep learning models. Designingand training a well-performing model is generally expensive. However, whenreleasing them, attackers may reverse engineer the models and pirate theirdesign. This paper, therefore, proposes deep learning obfuscation, aiming atobstructing attackers from pirating a deep learning model. In particular, wefocus on obfuscating convolutional neural networks (CNN), a widely employedtype of deep learning architectures for image recognition. Our approachobfuscates a CNN model eventually by simulating its feature extractor with ashallow and sequential convolutional block. To this end, we employ a recursivesimulation method and a joint training method to train the simulation network.The joint training method leverages both the intermediate knowledge generatedby a feature extractor and data labels to train a simulation network. In thisway, we can obtain an obfuscated model without accuracy loss. We have verifiedthe feasibility of our approach with three prevalent CNNs, i.e., GoogLeNet,ResNet, and DenseNet. Although these networks are very deep with tens orhundreds of layers, we can simulate them in a shallow network including onlyfive or seven convolutional layers. The obfuscated models are even moreefficient than the original models. Our obfuscation approach is very effectiveto protect the critical structure of a deep learning model from being exposedto attackers. Moreover, it can also thwart attackers from pirating the modelwith transfer learning or incremental learning techniques because the shallowsimulation network bears poor learning ability. To our best knowledge, thispaper serves as a first attempt to obfuscate deep learning models, which mayshed light on more future studies.

DeepObfuscation: Securing the Structure of Convolutional Neural Networks via Knowledge Distillation