- Rating
- 3(+1)
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper proposes J-Miner, an Executable Decision Compression (EDC) framework that extracts vocabulary-named variables from hidden representations and learns compact Boolean or linear decision rules. Across nine classification tasks and multiple model families, 16-variable rules achieve 88.8% average fidelity on six binary tasks. However, high behavioral fidelity does not establish causal faithfulness or semantic interpretability. Stronger causal validation and comparisons with competitive interpretability methods are needed to substantiate its claims.
Critical strengths
- Recovering interpretable and executable rules from fine-tuned LLM classifiers is important for model transparency and auditing.
- The method integrates internal feature extraction, variable selection, and rule learning into a unified approach.
- Experiments across multiple tasks and model families demonstrate consistent improvements over surface-word baselines.
- Standalone execution and the correction–harm decomposition provide insights into the relationship between internal-state reconstruction and behavioral fidelity.
Critical weaknesses
- High behavioral fidelity does not establish that the extracted rules capture the classifier's actual decision mechanism. The deletion experiments provide only indirect causal evidence.
- Comparisons with established rule-extraction and concept-based methods are limited. Vanilla logit-lens and SAE features also achieve higher fidelity than J-Lens.
- Vocabulary-named variables are not necessarily semantically meaningful or consistent across inputs.
- The standalone reader improves prediction fidelity despite lower concept-state reconstruction accuracy, with performance comparable to conventional knowledge distillation.
- Robustness under distribution shifts, different fine-tuning seeds, and larger models remains unexplored.
Questions for authors
- Can the authors provide stronger causal interventions to validate the recovered decision rules?
- How does J-Miner compare with established rule-extraction and concept-based methods under matched complexity?
- What measurable interpretability advantages does J-Lens offer over vanilla logit-lens and SAE features?
- How is the semantic consistency of vocabulary-named variables validated?
- Why does standalone execution improve behavioral fidelity while reducing concept-state fidelity?
- How robust are the recovered rules across distribution shifts, training seeds, and larger models?