# Knockoff Nets: Stealing Functionality of Black-Box Models

Source: [Knockoff Nets: Stealing Functionality of Black-Box Models](<https://www.sota2.com/research/paper/68fa65321e209131cea08d3a>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-25T21:20:52.699Z

Authors: Tribhuvanesh Orekondy, Bernt Schiele, Mario Fritz

Published: 2018-12-06T19:34:33.000Z

## Abstract

Machine Learning (ML) models are increasingly deployed in the wild to perform a wide range of tasks. In this work, we ask to what extent can an adversary steal functionality of such "victim" models based solely on blackbox interactions: image in, predictions out. In contrast to prior work, we present an adversary lacking knowledge of train/test data used by the model, its internals, and semantics over model outputs. We formulate model functionality stealing as a two-step approach: (i) querying a set of input images to the blackbox model to obtain predictions; and (ii) training a "knockoff" with queried image-prediction pairs. We make multiple remarkable observations: (a) querying random images from a different distribution than that of the blackbox training data results in a well-performing knockoff; (b) this is possible even when the knockoff is represented using a different architecture; and (c) our reinforcement learning approach additionally improves query sample efficiency in certain settings and provides performance gains. We validate model functionality stealing on a range of datasets and tasks, as well as on a popular image analysis API where we create a reasonable knockoff for as little as $30.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Image Classification on SVHN (test) (Standard Metrics)](<https://www.sota2.com/research/sota/image-classification-on-svhn-test-standard-metrics.md>): Image Classification; SVHN (test); Top-1 Accuracy: 83.37.

- [Image Classification on Tiny-ImageNet (Top-1/Top-5 Accuracy)](<https://www.sota2.com/research/sota/image-classification-on-tiny-imagenet-top-1-top-5-accuracy.md>): Image Classification; Tiny-ImageNet; Top-1 Accuracy: 50.22.

- [Image Classification on MNIST (test)](<https://www.sota2.com/research/sota/image-classification-on-mnist-test.md>): Image Classification; MNIST (test); Accuracy: 98.81.

- [Image Classification on ImageNet-1K (Top-1 Accuracy)](<https://www.sota2.com/research/sota/image-classification-on-imagenet-1k-top-1-accuracy.md>): Image Classification; ImageNet-1K; Top-1 Acc: 57.43.

- [Image Classification on CIFAR-100 (test)](<https://www.sota2.com/research/sota/image-classification-on-cifar-100-test.md>): Image Classification; CIFAR-100 (test); Accuracy: not recorded.

- [Image Classification on CIFAR-100](<https://www.sota2.com/research/sota/image-classification-on-cifar-100.md>): Image Classification; CIFAR-100; Accuracy: not recorded.

- [Image Classification on TinyImageNet (test)](<https://www.sota2.com/research/sota/image-classification-on-tinyimagenet-test.md>): Image Classification; TinyImageNet (test); Accuracy: not recorded.

- [Image Classification on CIFAR-10 (Standard Accuracy)](<https://www.sota2.com/research/sota/image-classification-on-cifar-10-standard-accuracy.md>): Image Classification; CIFAR-10; Accuracy: not recorded.

- [Image Classification on MNIST](<https://www.sota2.com/research/sota/image-classification-on-mnist.md>): Image Classification; MNIST; Accuracy: not recorded.

- [Model Stealing on CIFAR-10 (test)](<https://www.sota2.com/research/sota/model-stealing-on-cifar-10-test.md>): Model Stealing; CIFAR-10 (test); Synthetic Accuracy (Data-Free): not recorded.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/1812.02766>)

[PDF](<https://arxiv.org/pdf/1812.02766.pdf>)
