adversarial-friend

CS_5523_Final_Project

This repo contains code and files of final project in CSCI 5523: Data Mining

Getting Started

To import AdversarialFriend API, use the following pip command: pip install adversarial-friend

Team Members

Yangyang Li li002252@umn.edu

Jacob Isler isle0011@umn.edu

Mario Serrafero serra082@umn.edu

Project Abtract

We will explore parameter interpretation techniques for computer vision tasks, in order to get a deeper understanding of how to deal with real image data. Specifically speaking, the models to be used include generalized linear models like Logistic Regression, as well as more powerful techniques such as convolutional neural networks. What is more, we will document and implement these algorithms to extract the learned concepts that our models detect in input images. Finally, we will use the information, along with automatic input optimization, to programmatically generate adversarial examples using model-agnostic gradient-based methods that trick the learned models into misclassifying input images.

For full project report, go to: [add link])()

Reference

[1] C. Olah, A. Satyanarayan, I, Johnson, S. Carter, L. Schubert, K. Ye, A. Mordvinstev, “The Building Blocks of Interpretability” Distill, 6-Mar-2018. Online [Accessed: 1-Dec-2020]

[2] J. Johnson, EECS 498-007 / 598-005 Deep Learning for Computer Vision, University of Michigan, 10-Aug-2020. Online [Accessed: 1-Dec-2020]

[3] A. Kurakin, I. Goodfellow, S. Bengio, “Adversarial Examples in the Physical World” ICLR, 8-Jul-2016. Online [Accessed: 1-Dec-2020]

License

This is free and unencumbered software released into the public domain.

Anyone is free to copy, modify, publish, use, compile, sell, or distribute this software, either in source code form or as a compiled binary, for any purpose, commercial or non-commercial, and by any means.