📚 References

🔬 Research Outputs

• Bridging Traditional Explainability Methods and Multimodal Multilingual Models

• SGPA: Spectrogram-Guided Phonetic Alignment for Feasible Shapley Value Explanations

• mllm-shap Zenodo release

🗂️ Repository Resources

• Main repository

• Experiments folder

• Examples folder

• Package tests

⚠️ Notes

Approximation methods can become unstable for very small sample budgets or short token sequences. For production workloads, validate approximation quality against stronger baselines first.

Warning

Always validate Monte Carlo-style settings on representative data before running large sweeps.