[EXE] Robustness of Machine Learning Models - Githubissues

gimseng / 99-ML-Learning-Projects

A list of 99 machine learning projects for anyone interested to learn from coding and building projects

MIT License

575 stars 173 forks source link

[EXE] Robustness of Machine Learning Models #150

Open hash-ir opened 3 years ago

hash-ir commented 3 years ago

Learning Goals

Create adversarial examples of input data using gradient descent (e.g Fast Gradient Attack)
Train and visualize predictions of semantically same inputs (original and perturbed)
Quantitative evaluation of the accuracy against training time (including adversary creation)

Exercise Statement

Create adversarial examples and train a CNN on clean and perturbed data using PyTorch

Prerequisites

Basics of creating and training neural networks in PyTorch
Projected gradient descent, an idea about norm balls (L1, L2, L-inf)

Data source/summary:

Could be any image data (better explained and visualized with MNIST)

(Optional) Suggest/Propose Solutions

I have the solution using PyTorch, will be happy to create a pull request to include the exercise statement/solution