Safeguarding Machine Learning

adversarial-nik commented 5 years ago

As a security researcher, I would like to open a discussion to make Machine Learning practices secure over ubiquity. We will discuss "not so commonly known" vulnerabilities in machine learning applications with some demos and real world incidents. Also, the mitigation if any. Outline of talk -

The pipeline AKA attack surface:

We will start with an introduction to Machine Learning pipeline and then discuss how an attacker can abuse its access to different steps in pipeline.

Adversarial Learning Attacks:

This is the only widely acknowledged attack by data science community. So we will dive straight into a short demo and discuss a case study on how a Russian company used it to their own advantage.

Model stealing Attack:

This can be a direct attack on revenue generation of target organisation
Offline Model Stealing
- A demo on how deployed models can be stolen by attackers when its deployed on locally
- The attack will be demonstrated on a real world application
online model stealing
- Targeting models deployed on cloud
- Discussion on PRADA and PoC of GDALR

Data poisoning and Model Skewing attacks:

How feedback to model from user can be abused to skew the predictions of model
Demo and mitigation

Model Inversion:

Inferring sensitive training data used to train the model when given white box access to the model.
Possible mitigation

This will be a 45-50 minutes presentation.

Thanks!

mayankskb commented 4 years ago

Hi,

We would like to have your talk on Jan 18, 2020. Please confirm your availability for the same.

Thanks Mayank Mishra

adversarial-nik commented 4 years ago

Hey Mayank, I will be available on 18th Jan. May I know how much time will be allocated for this talk so that I can think about delivering demos.

mayankskb commented 4 years ago

Hi,

As we had your lightning talk in January, Can we move with this talk proposal for April meetup. Mostly that will be a virtual session via slack or youtube live.

Please let me know your availability for April 11, 2020.

Thanks Mayank Mishra

adversarial-nik commented 4 years ago

Sure Mayank,

I will be available on 11th April for live session.

mayankskb commented 4 years ago

Hi,

Please let us know if we can have your talk for August 15, virtual session.

Thanks Mayank Mishra

adversarial-nik commented 4 years ago

Hi Mayank,

Not sure at what time you are planning for the virtual session. But I will be available after 3PM on 15th August.

mayankskb commented 4 years ago

Well we are planning to keep it for 7pm. Lets us know if this suits to you.

adversarial-nik commented 4 years ago

sure, lets do it.

mayankskb commented 4 years ago

Ok and any changes to the theme and line up that you have mentioned in the description or to the title.. If there is any please let me know.

Also if you can send me a headshot for flyer at mayank1996[dot]skb[at]gmail[dot]com

mayankskb commented 4 years ago

Accepting it for August 15, 7pm youtube live event

PyDataPune / Talks