LLE

Code implementation for paper Lifelong Explainer for Lifelong Learners at EMNLP 2021, by Xuelin Situ, Sameen Maruf, Ingrid Zukerman, Cecile Paris and Reza Haffari.

Requirements and Installation

• Python version >= 3.6.8
• PyTorch version >= 1.7.0
• HuggingFace transformers version >= 1.2.0
• LIME >= 0.1.1.36
• shap == 0.29.3

Experiments (steps to replicate the results from the paper)

1. Train lifelong black-box classifier >> preprocess.train_lifelong_classifier.py

2. Collect explanations from different teachers >> preprocess.collect_teacher_explanations.py

3. Train LLE explainer (also refer to folder hyperparameters) >> lifelongexplanation.py

4. Faithfulness evaluation >> evaluation.compare_faithfulness.py

5. Stability evaluation >> evaluation.compare_stability.py

6. Efficiency evaluation >> evaluation.compare_efficiency.py

7. Ablation Study of Experience Replay:

   • Faithfulness >> evaluation.ablation_faithfulness.py
   • Stability >> evaluation.ablation_stability.py