Using Stratified Sampling to Improve LIME Image Explanations
Abstract: We investigate the use of a stratified sampling approach for LIME Image, a popular model-agnostic explainable AI method for computer vision tasks, in order to reduce the artifacts generated by typical Monte Carlo sampling. Such artifacts are due to the undersampling of the dependent variable in the synthetic neighborhood around the image being explained, which may result in inadequate explanations due to the impossibility of fitting a linear regressor on the sampled data. We then highlight a connection with the Shapley theory, where similar arguments about undersampling and sample relevance were suggested in the past. We derive all the formulas and adjustment factors required for an unbiased stratified sampling estimator. Experiments show the efficacy of the proposed approach.
- ImageNet Object Localization Challenge. https://kaggle.com/competitions/imagenet-object-localization-challenge.
- Benchmarking and survey of explanation methods for black box models. Data Mining and Knowledge Discovery, 1–60.
- Using sample survey weights in multiple regression analyses of stratified samples. Journal of the American Statistical Association, 78(383): 535–543.
- Explaining the explainer: A first theoretical analysis of LIME. In Int. Conf. on artificial intelligence and statistics, 1287–1296. PMLR.
- What does LIME really see in images? In Int. Conf. on machine learning, 3620–3629. PMLR.
- A survey of methods for explaining black box models. ACM computing surveys (CSUR), 51(5): 1–42.
- Haberman, S. J. 1975. How much do Gauss-Markov and least square estimates differ? A coordinate-free approach. The Annals of Statistics, 982–990.
- Deep residual learning for image recognition. In Proceedings of the IEEE Conf. on computer vision and pattern recognition, 770–778.
- Konijn, H. S. 1962. Regression analysis in sample surveys. Journal of the American Statistical Association, 57(299): 590–606.
- Explaining the black-box model: A survey of local interpretation methods for deep neural networks. Neurocomputing, 419: 168–182.
- A unified approach to interpreting model predictions. Advances in neural information processing systems, 30.
- An analysis of LIME for text data. In Int. Conf. on Artificial Intelligence and Statistics, 3493–3501. PMLR.
- Sampling permutations for Shapley value estimation. The Journal of Machine Learning Research, 23(1): 2082–2127.
- Variations on the Shapley value. Handbook of game theory with economic applications, 3: 2055–2076.
- Towards more stable LIME for explainable AI. In 2022 Int. Symposium on Intelligent Signal Processing and Communication Systems (ISPACS), 1–4. IEEE.
- A multilinear sampling algorithm to estimate Shapley values. In 25th Int. Conf. on Pattern Recognition (ICPR), 7992–7999. IEEE.
- Owen, G. 1972. Multilinear extensions of games. Management Science, 18(5-part-2): 64–79.
- Rao, T. J. 1977. Optimum Allocation of Sample Size and Prior Distributions: A Review. Int. Statistical Review, 45(2): 173–179.
- Why should I trust you? Explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD int. Conf., 1135–1144.
- The Shapley value in machine learning. IJCAI, arXiv:2202.05594.
- REVEL Framework to Measure Local Linear Explanations for Black-Box Models: Deep Learning Image Classification Case Study. Int. Journal of Intelligent Systems, 2023.
- ALIME: Autoencoder based approach for local interpretability. In Intelligent Data Engineering and Automated Learning (IDEAL), 454–463. Springer.
- An extension of LIME with improvement of interpretability and fidelity. arXiv preprint arXiv:2004.12277.
- Quick shift and kernel methods for mode seeking. In Computer Vision–ECCV 2008, Proceedings, Part IV 10, 705–718. Springer.
- Explainable image classification with evidence counterfactual. Pattern Analysis and Applications, 25(2): 315–335.
- Statistical stability indices for LIME: Obtaining reliable explanations for machine learning models. Journal of the Operational Research Society, 73(1): 91–101.
- ”Why Should You Trust My Explanation?” Understanding Uncertainty in LIME Explanations. ICML.
Paper Prompts
Sign up for free to create and run prompts on this paper using GPT-5.
Top Community Prompts
Collections
Sign up for free to add this paper to one or more collections.
