Algorithms that remember: model inversion attacks and data protection law

Note: Sign-ups to this event are managed through Skills Matter. Please sign up at: https://skillsmatter.com/meetups/12879-algorithms-that-remember-model-inversion-attacks-and-data-protection-law

In the presence of model inversion and membership inference attacks, should we consider models as personal data and apply the same protections as we would to the underlying data? What would it mean in practice to apply data protection regulations to a model in this circumstance?

In research connecting the development of statistical models and the regulatory frameworks that our work is done within, this month’s paper explores these questions and looks at what lessons we can draw from this exercise.

The main body of the discussion will be around the paper below

• Paper - Algorithms that remember: model inversion attacks and data protection law - Veale, Binns, Edwards (2018)
• https://arxiv.org/pdf/1807.04644.pdf

There is also a blog by one of the authors which summarises the issues

• Blog - https://ai-auditingframework.blogspot.com/2019/09/privacy-attacks-on-ai-models.html

Further materials (not necessary for the discussion, but go into more technical aspects of how these attacks might work)

• Video on Membership Inference attacks - https://www.youtube.com/watch?v=rDm1n2gceJY
• Paper on Model Inversion attack - https://www.cs.cmu.edu/~mfredrik/papers/fjr2015ccs.pdf

—

A note about the Journal Club format:

1. The sessions usually start with a 5-10 minute introduction to the paper by the topic volunteer, followed by splitting into smaller groups to discuss the paper and other materials. We finish the session by coming together for about 15 minutes to discuss what we have learned as a group and ask questions around the room.
2. There is no speaker at Journal Club. One of the community has volunteered their time to suggest the topic and start the session, but most of the discussion comes from within the groups.
3. You will get more benefit from the session if you read the paper or other materials in advance. We try to provide (where we can find them) accompanying blog posts, relevant code and other summaries of the topic to serve as entry points.
4. If you don't have time to do much preparation, please come anyway. You will probably have something to contribute, and even if you just end up following the other discussions, you can still learn a lot.
5. It's OK just to read the blog post or watch the video :)
6. We don't have spare copies of the paper during the session, so please print out your own if you want a hard copy for discussion. For digital copies, you are welcome to use your laptops/tablets/phones during the session.