nps1ngh / adversarial-bert-german-attacks-defense

White-Box Attacks on BERT in the German Language with Explicit and Implicit Character Level Defense

White-Box Attacks on BERT in the German Language with Explicit and Implicit Character Level Defense

Links to code sections

• BERT Finetunig
• Attacks: Whitebox Baseline, Character-level and Word-level
• Defenses: Explicit Character-level and Abstain label training
  Datasets:
• Germeval 2021 Task 1: Toxic Comment Classification
• HASOC (2019) German Language: Sub Task 1, Hate Speech Classification