Machine Learning Safety

This repository includes the code (in Chapter-X/) and the draft textbook (file "Machine_Learning_Safety.pdf"). It is used in "COMP219: Advanced Artificial Intelligence" at the Univesity of Liverpool for second year undergraduate students.

Table of Contents:

Part 1: Safety Properties

Part 2: Safety Threats

Part 3: Safety Solutions

Part 4: Extended Safety Solutions

Part 5: Appendix: Mathematical Foundations and Competition

Installation

conda installation

windows https://conda.io/projects/conda/en/latest/user-guide/install/windows.html

Linux: https://docs.conda.io/projects/conda/en/latest/user-guide/install/linux.html

macOS: https://docs.conda.io/projects/conda/en/latest/user-guide/install/macos.html

conda env setup

First of all, you can set up a conda environment (Note that you do not need to set up conda in the lab)

conda create --name aisafety python==3.7
conda activate aisafety

This should be followed by installing software dependencies:

conda install -c pandas numpy matplotlib tensorflow scikit-learn pytorch torchvision

IDE (VScode/PyCharm) and JupyterNotebook with conda

1. Visual Studio Code (VScode) is a free cross-platform source code editor. The Python for Visual Studio Code extension allows VSC to connect to Python distributions installed on your computer. Reference: https://docs.anaconda.com/anaconda/user-guide/tasks/integration/python-vsc/
2. Using PyCharm. PyCharm is an IDE that integrates with IPython Notebook, has an interactive Python console, and supports Anaconda as well as multiple scientific packages. Reference: https://docs.anaconda.com/anaconda/user-guide/tasks/pycharm/
3. If you want to use Jupyter Notebook to code, use below command:

• install jupyter

conda install jupyter

• open jupyter notebook

jupyter notebook