Reproduce “Faithful Explanations of Black-box NLP Models Using LLM-generated Counterfactuals” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yair Gat, Nitay Calderon, Amir Feder, Alexander Chapanin, Amit Sharma, Roi Reichart
Researchers developed methods for faithful explanations of black-box NLP models by leveraging Large Language Models (LLMs) to generate counterfactuals. The work introduces both direct LLM-generated counterfactuals and an LLM-guided causal representation learning for efficient matching, achieving state-of-the-art performance in explanation error reduction and demonstrating LLMs' utility for creating new causal explanation benchmarks.
Key insights
- LLMs can effectively generate high-quality counterfactuals by performing complex causal interventions, making them powerful tools for both explanation and benchmark creation.
- Learning a causal embedding space significantly improves the efficiency and accuracy of finding approximate counterfactuals for black-box NLP models.
- The "Order-faithfulness" criterion provides a strong theoretical foundation for prioritizing causal, counterfactual-based explanations over non-causal alternatives.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.