Skip to content

Latest commit

 

History

History
41 lines (38 loc) · 2.04 KB

README.md

File metadata and controls

41 lines (38 loc) · 2.04 KB

Controlling Information Leakage in Concept Bottleneck Models with Trees

MSc Thesis Project by Angelos Ragkousis

Table of Contents

Abstract

In this work, we propose a new method to train a CBM while retaining Leakage Inspection and Control properties using Decision Trees. This approach can be used for both sequentially and jointly trained CBMs, and is also tested in a Mix- ture of Experts Framework. Our method uses Information Leakage to extend the Decision Paths of a classifier which have missing concept information. It allows for Inspection and Control for specific subsets of data, improves the task accuracy for these subsets, and provides more meaningful explanations.

Installation

Steps to set up the environment and install dependencies:

  1. Clone the repository: 
    git clone https://github.com/AggelosRag/AR-Imperial-Thesis.git
cd your-repo
  2. Create and activate a virtual environment: 
    python -m venv env
source env/bin/activate  # On Windows, use `env\Scripts\activate`
  3. Install dependencies (not added yet, packages need to be installed manually): 
    pip install -r requirements.txt

Usage

How to run an experiment: The train.py script should be called, using the appropriate configuration file from the ./configs folder. Two examples are given below:

  • To train a Vanilla CBM model on the Morpho-MNIST dataset with independent training, use the following configuration file (change hyperparameters as needed): 
    python train.py --config ./configs/CBM/Independent/No_Regularisation/mnist_full.json
  • To train the MCBM-Seq algorithm on the Morpho-MNIST dataset, use the following configuration file (change hyperparameters as needed): 
    python train.py --config ./configs/LeakageInspection/mnist.json

License

Project license: This project is licensed under the MIT License - see the LICENSE file for details.