Skip to content

Latest commit

 

History

History
16 lines (11 loc) · 1.44 KB

README.md

File metadata and controls

16 lines (11 loc) · 1.44 KB

Collection of Training-Data Fault Mitigation (TDFM) Techniques

This repository contains our modified implementations of the TDFM approaches described in our DSN'22 paper, as well as the experimental results - The Fault in Our Data Stars: Studying Mitigation Techniques against Faulty Training Data in Machine Learning Applications.

We list the original implementers of these tools below - these are based on publicly available sources.

  1. Label Smoothing: Originally from LabelRelaxation. Our modified version in LabelRelaxation
  2. Label Correction: Originally from MLC. Our modified version in MLC
  3. Robust Loss: Originally from Active-Passive-Losses . Our modified version in Active-Passive-Losses
  4. Knowledge Distillation: KD
  5. Ensemble: NN-Ensemble

Supplementary Material

  1. The complete set of figures for all configurations, including multiple fault type injection and runtime analysis, is accessible here.
  2. The complete results in table form for all configurations is accessible here.