Project Option 3

Explainable Machine Learning

Model Interpretability and Visualization

Description

Study the visualization methods to explain the results of CNNs. Implement at least two visualization algorithms on at least two datasets: at least one toy dataset and at least one real-world dataset (CIFAR and ImageNet belong to this class). Compare and analyze the visualization results among methods.

Basic Readings

Explainable AI (XAI) with Class Maps, 2022.
Visualizing Backpropagation in Neural Network Training at Any Scale, 2022. HiPlot Github, parallel coordinate plots
Data Visualization in Python, 2021.
Exploratory Data Analysis in Python, 2021.
Deep learning doesn’t need to be a black box, 2021.
Visualizing Feature Maps using PyTorch, 2021.
揭密 AI 黑箱，開發者投入研究「可解釋的人工智慧」 2021.
Making Interactive Visualizations with Python Altair, 2021.
Guide to Interpretable Machine Learning - Techniques to dispel the black box myth of deep learning. Towards Data Science, 2020.
Interpretability in Machine Learning, Medium, 2020.
Is Explainable AI (xAI) the Next Step, or Just Hype? 2020.
Mengnan Du, Ninghao Liu, Xia Hu, "Techniques for Interpretable Machine Learning," Communications of the ACM, Vol. 63 No. 1, Pages 68-77, January 2020.
Decoding the Black Box: An Important Introduction to Interpretable Machine Learning Models in Python, 2019.
Python Libraries for Interpretable Machine Learning, 2019.
The great AI debate: Interpretability, Medium, 2019.
人工智慧新發展，可向人類解釋思考過程，2018.
MIT Lincoln Laboratory develops AI that shows its decision-making process, 2018
J. Yosinski, J. Clune, A. Nguyen, T. Fuchs, H. Lipson, "Understanding Neural Networks Through Deep Visualization," ICML 2015.
WELCOME TO PYTORCH TUTORIALS, PyTorch Web Site

Two Sub-topics

Machine Learning Visualization, CNN Visualization

ML VISUALIZATION

Generate Publication-Ready Plots Using Seaborn Library, 2020.
- Part-1. Different types of plots using seaborn
- Part-2. Facet, Pair and Joint plots using seaborn
Visdom (GitHub) [PyTorch], Facebook Research
YellowBrick, ELI5, LIME, MLxtend.
Machine Learning Visualization, 2020.
Python Libraries for Interpretable Machine Learning, 2019.
LIME vs. SHAP: Which is Better for Explaining Machine Learning Models? 2020

CNN VISUALIZATION

Explainable MNIST classification, 2020.
- Part 1 : dissection of a ConvNet. Explaining the decisions of a ConvNet in classifying MNIST images
- Part 2: building a ConvNet by hand (Not Yet Completed)
Papers about interpretable CNN (GitHub)
Awesome Interpretable Machine Learning (GitHub)
GWU_data_mining/10_model_interpretability (GitHub)