aime-xai 1.1.0

Approximate Inverse Model Explanations (AIME): unified global/local importance for XAI

**AIME:**Approximate Inverse Model Explanations

The AIME methodology is detailed in the paper available at The AIME methodology is detailed in the paper available at https://ieeexplore.ieee.org/document/10247033. AIME is proposed to address the challenges faced by existing methods in providing intuitive explanations for black-box models. AIME offers unified global and local feature importance by deriving approximate inverse operators for black-box models. It introduces a representative instance similarity distribution plot, aiding comprehension of the predictive behavior of the model and target dataset.

Features

• Unified Global and Local Feature Importance: AIME derives approximate inverse operators for black-box models, offering insights into both global and local feature importance.

• Representative Instance Similarity Distribution Plot: This feature aids in understanding the predictive behavior of the model and the target dataset, illustrating the relationship between different predictions.
• Effective Across Diverse Data Types: AIME has been tested and proven effective across various data types, including tabular data, handwritten digit images, and text data.
• Intuitive Explanations: AIME's explanations are simpler and more intuitive than those generated by well-established methods like LIME and SHAP.
• Visualization of Similarity Distribution: AIME visualizes the similarity distribution with the target dataset, providing insights into the relationship between different predictions.

License

AIME is dual-licensed under the The 2-Clause BSD License and the Commercial License. Apply the The 2-Clause BSD License only for academic or research purposes, and apply Commercial License for commercial and other purposes. You can choose which one to use.

Commercial License

For those interested in Commercial License, a licensing fee may be required. Please contact us for more details at: Email: [e-mail]

Installation

pip install aime-xai

PCTAIME :PCAIME: Principal Component Analysis-Enhanced Approximate Inverse Model Explanations Through Dimensional Decomposition and Expansion

Please refer to PCAIME.ipynb when implementing.

HuberAIME :

AIME is based on the least squares method, so it has the disadvantage of being weak against outliers. If you want to include data with outliers, please use it as “AIME (use_huber=True)”.

Citation

If you use this software for research or other purposes, please cite the following paper.

@ARTICLE{10247033,
author={Nakanishi, Takafumi},
journal={IEEE Access}, 
  title={Approximate Inverse Model Explanations (AIME): Unveiling Local and Global Insights in Machine Learning Models}, 
  year={2023},
  volume={11},
  number={},
  pages={101020-101044},
 doi={10.1109/ACCESS.2023.3314336}}

@ARTICLE{10979913,
  author={Nakanishi, Takafumi},
  journal={IEEE Access}, 
  title={HuberAIME: A Robust Approach to Explainable AI in the Presence of Outliers}, 
  year={2025},
  volume={13},
  number={},
  pages={76796-76810},
  keywords={Computational modeling;Robustness;Predictive models;Estimation;Accuracy;Computational efficiency;Atmospheric modeling;Explainable AI;Closed box;Iterative methods;Approximate inverse model explanations;explainable AI;global feature importance;Huber loss;iterative reweighted least squares;model-agnostic explanations;outliers;robustness},
  doi={10.1109/ACCESS.2025.3565279}}

@ARTICLE{10648696,
  author={Nakanishi, Takafumi},
  journal={IEEE Access}, 
  title={PCAIME: Principal Component Analysis-Enhanced Approximate Inverse Model Explanations Through Dimensional Decomposition and Expansion}, 
  year={2024},
  volume={12},
  number={},
  pages={121093-121113},
  keywords={Correlation;Analytical models;Principal component analysis;Feature extraction;Artificial intelligence;Estimation;Dimensionality reduction;Explainable AI;Approximation methods;Approximate inverse model explanation;explainable artificial intelligence;feature correlation;feature importance;model explanation;principal component analysis;principal component analysis-enhanced approximate inverse model explanation},
  doi={10.1109/ACCESS.2024.3450299}}