Pitting machine vision models against adversarial attacks.
Welcome to the Adversarial Vision Challenge, one of the official challenges in the NIPS 2018 competition track. In this competition you can take on the role of an attacker or a defender (or both). As a defender you are trying to build a visual object classifier that is as robust to image perturbations as possible. As an attacker, your task is to find the smallest possible image perturbations that will fool a classifier.
The overall goal of this challenge is to facilitate measurable progress towards robust machine vision models and more generally applicable adversarial attacks. As of right now, modern machine vision algorithms are extremely susceptible to small and almost imperceptible perturbations of their inputs (so-called adversarial examples). This property reveals an astonishing difference in the information processing of humans and machines and raises security concerns for many deployed machine vision systems like autonomous cars. Improving the robustness of vision algorithms is thus important to close the gap between human and machine perception and to enable safety-critical applications.
There will be three tracks in which you and your team can compete:
Defense track: Your task is to build and train a robust model on tiny ImageNet. The attacks will try to find small image perturbations that change the prediction of your model to the wrong class. The larger these perturbations are the better is your score (the exact scoring formula will be published soon).
Untargeted attacks: In this track you build an attack algorithm that breaks the defenses. For each model and each given image your attack tries to find the smallest perturbation that makes the model predict a wrong class label (so-called adversarial perturbations). Your attack will be able to craft model-specific adversarials by asking the model for its prediction on self-defined inputs (up to 1000 times / image). The smaller the adversarial perturbations are that your attack finds (on average), the better is your score (the exact scoring formula will be published soon).
Targeted attacks: This track is very similar to the untargeted attack track. The only difference is that here an adversarial perturbation is not defined as making the model predict any wrong label but it has to get the model to predict a particular (wrong) label.
Models are scored as follows (higher is better):
- Let M be the model and S be the set of samples.
- We apply the five best untargeted attacks on M for each sample in S.
- For each sample we record the minimum adversarial L2 distance (MAD) across the attacks.
- If a model misclassifies a sample then the minimum adversarial distance is registered as zero for this sample.
- The final model score is the median MAD across all samples.
- The higher the score, the better.
Attacks are scored as follows (lower is better):
- Let A be the attack and S be the set of samples.
- We apply attack A against the best five models for each sample in S.
- If an attack fails to produce a (targeted) adversarial for a given sample, then we register a worst case distance (distance of the sample to a uniform grey image).
- The final attack score is the median L2 distance across samples.
The top-5 models and attacks against which submissions are evaluated are fixed for two weeks at a time after which we evaluate all current submissions to determine the new top-5 models and attacks for the upcoming two weeks.
* June 25th, 2018 : Challenge begins
* November 1st : Final submission date
* November 15th : Winners Announced
The organizing team comes from multiple groups — University of Tübingen, Google Brain, EPFL and Pennsylvania State University.
To be announced
A starter kit has currently being prepared and will explain all the nuts and bolts required to get started in the challenge. Stay tuned!
- Gitter Channel : crowdAI/nips-2018-adversarial-vision-challenge
- Discussion Forum : https://www.crowdai.org/challenges/nips-2018-adversarial-vision-challenge/topics
We strongly encourage you to use the public channels mentioned above for communications between the participants and the organizers. In extreme cases, if there are any queries or comments that you would like to make using a private communication channel, then you can send us an email at :