Attacking Machine Learning Models

The Implications

Doesn’t look like anything to me!

Adversarial Attacks

What are the reasons for such a vulnerability? How can we come up with adversarial examples? How can we defend our models against such attacks?

So, why does such a vulnerability exist?

Ref: Machine Learning at Berkeley

Generating Adversarial Examples

Fast Gradient Sign Method

Basic Iterative Method

Targeted Fast Gradient Sign Method

What can we do to defend our models?

Hide your gradients!

Be prepared with Adversarial Training

Acknowledging Ignorance

How we assign null labels to examples that are perturbed with calculated noise.

Conclusion

--

--

--

Interested in AI Safety Research

Love podcasts or audiobooks? Learn on the go with our new app.

Recommended from Medium

A Simple Guide to Using Keras Pretrained Models

Deep Learning based Pitch Detection (CNN, LSTM)

Implementing word2vec in PyTorch (skip-gram model)

Everything you need to know about MLOps for your company

A Beginner’s guide to Supervised Learning

Depicts a classroom setup resembling to concept of supervised learning.

Real-Time Analytics: COVID-19 Weekly Roundup February 3rd, 2021

Please Vote Uniqtech Publication Pipeline

Subsurface Salt Interpretation Automation — Enhancing ML workflow for semantic segmentation

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Sai Sasank

Sai Sasank

Interested in AI Safety Research

More from Medium

Neural Gas Simplified

What is Linear Discriminant Analysis ?|Assumptions

Deep Learning approach to skin cancer detection.

Machine learning