backdooring-semantic-loss-with-wanet

Benchmarking the robustness of Neuro Symbolic Models against Backdoor Attacks using Semantic Loss and WaNet

https://github.com/francescohamar/backdooring-semantic-loss-with-wanet

Science Score: 44.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • CITATION.cff file
    Found CITATION.cff file
  • codemeta.json file
    Found codemeta.json file
  • .zenodo.json file
    Found .zenodo.json file
  • DOI references
  • Academic publication links
  • Academic email domains
  • Institutional organization owner
  • JOSS paper metadata
  • Scientific vocabulary similarity
    Low similarity (8.5%) to scientific vocabulary

Keywords

tudelft-cse-research-project
Last synced: 6 months ago · JSON representation ·

Repository

Benchmarking the robustness of Neuro Symbolic Models against Backdoor Attacks using Semantic Loss and WaNet

Basic Info
  • Host: GitHub
  • Owner: FrancescoHamar
  • Language: Python
  • Default Branch: main
  • Homepage:
  • Size: 17.5 MB
Statistics
  • Stars: 0
  • Watchers: 1
  • Forks: 0
  • Open Issues: 0
  • Releases: 0
Topics
tudelft-cse-research-project
Created 10 months ago · Last pushed 8 months ago
Metadata Files
Readme Citation

README.md

Backdooring Semantic Loss with WaNet

This repository is part of the bachelor thesis concluding the BSc Computer Science & Engineering, for the Research Project 2025 of TU Delft.

Project Directory

Constraint sets

All constraint sets used in the form of a constraints.txt can be found in the data correlation and constraint sets directory.

Output

Almost all output mentioned in the paper is also shown in the repository, separated into three folder by poison rate.

In each folder there are 4 subfolder. Backdoor and Clean give example images of how the classification was going at the last epoch. The general folder graphs about the performance, and the full folder has raw output with arrays of values, all labeled in a txt file.

Scripts

During the project some scripts were used for image generation and data processing, some of these were saved and can be found in the scripts folder.

The model

The main runnable code can be found in the Semantic Loss Model folder. This folder also contains another README.md with further information.

About the results

Some results can be observed from the figures in the appropriate folder, but of course for further information please refer to the full paper.

Contact

For any question please refer to F.Hamar@student.tudelft.nl or francesco.hamar@gmail.com

Owner

  • Name: Francesco Hamar
  • Login: FrancescoHamar
  • Kind: user
  • Location: Delft

🏫 Studying Computer Science and Engineering at TU Delft.

Citation (CITATION.cff)

cff-version: 1.2.0
message: "If you use this software, please cite it as below."
authors:
  - family-names: Hamar
    given-names: Francesco
title: "Evaluating the Robustness of Neuro-Symbolic Networks Against Backdoor Threats with WaNet and Semantic Loss"
version: 1.0.0
date-released: 2025-06-19

GitHub Events

Total
  • Push event: 1
Last Year
  • Push event: 1