2952Q_FinalProject : Robust Estimation for the Erdős-Rényi Model

This repository contains the code and resources for the final project of the class CSCI2952Q: Robust Algorithms in Machine Learning. The project focuses on the robust estimation of the edge probability (p) in the Erdős-Rényi random graph model under adversarial perturbation of vertices.

Abstract

We define a new class of adversarial models, the ((q, \epsilon))-adversarial model, and we present novel algorithms: mean_adjusted_mean and the variance_method.

Repository Layout

• code/: Contains the implementation of various algorithms and experiments.
  • algos.py: Implementation of robust estimation algorithms.
  • basic.py: Basic graph operations and utilities.
  • figures.ipynb: Code for figures 2,3,4,5 in the report
  • figures2.ipynb: Code for figure 6 in the report
  • var_adversary.py: Adversarial perturbation algorithms for the variance method.
  • surveys/: Experiments in other methods.
• report/: Contains the LaTeX source files for the final report.
  • final.tex: Main LaTeX file for the report.
  • egbib.bib: Bibliography file.
  • img/: Directory containing images used in the report.
• presentation/: Contains the presentation slides for the project.

Paper

Presentation