안녕하세요! Robert 입니다.
이번 포스트 에서는 Explainable AI (XAI) vs. Interpretable AI 에 대한 내용을 다뤄보고자 합니다.
주로 참고한 포스트는 Richard Gall 님 께서 KD Nuggets 에 올려주신 Machine Learning Explainability vs Interpretability: Two concepts that could help restore trust in AI 입니다. * 원문도 읽어보시는 것을 추천드립니다!
Explainable AI? Interpretable AI? 에 대한 개인적인 호기심을 위주로 조사해봤습니다.
1. AI (Deep learning) Explainability 가 필요한 이유
2. Explainability vs. Interpretability
3. 이를 해결하기 위해 현재 진행중인 연구
Contents
1. Why do we need Explainable AI (XAI)?
2. Explainability vs. Interpretability
3. What kind of Research?
Why do we need XAI?
In some fields, failure is not an option: even a momentarily dysfunctional computer vision algorithm in autonomous vehicle easily leads to fatality. In the medical field, clearly human lives are on the line - Erico Tjoa
Explainable AI vs. Interpretable AI [5]
- Explainable AI: The knowledge of both what a node represents and its importance to the model's performance
- Interpretable AI: The ability to determine cause and effect from a machine learning model
Explainable Machine Learning/Deep Learning
- LOCO (Leave One Column Out)
- Permutation Impact (PI)
- LIME (Local Interpretable Model-agnostic Explanations)
You can easily find the explanation as well as python codes about LIME here
- CAM (Class Activation Map)
- LRP (Layer-wise Relevance Propagation)
- DEEPLIFT
XAI for Medicine
Personally, the most interesting topic is explainable AI for medicine. Considering the failure of IBM Watson, it is necessary to conduct advanced research on XAI for the further development of bench-to-bedside translational AI applications. Luckily, there is a review paper on XAI for medicine and we can check a research trend in this field.
By applying the same categorization to interpretability in medical research, it is hoped that (1) clinicians and practitioners can subsequently approach these methods with caution.(2) insights into interpretability will be born with more considerations for medical practices, and (3) initiatives to push forward data-bassed, mathematically- and technically grounded medical education are encouraged. - Erico Tjoa, et al.
Table 1, 2 in this paper, represents a list of journal papers arranged according to the interpretability methods used. Methods are categorized by HSI (Human Study on Interpretability), ANN (Artificial Neural Network), and Mechanism (Decomposition, Sensitivity, etc.)
Reference
[1] Budzik, J. (2018). Four Approaches to Explaining AI and Machine Learning. KDnuggets. https://www.kdnuggets.com/2018/12/four-approaches-ai-machine-learning.html
[2] Tjoa, E., & Guan, C. (2020). A Survey on Explainable Artificial Intelligence (XAI): Toward Medical XAI. IEEE Transactions on Neural Networks and Learning Systems, 1–21. https://doi.org/10.1109/tnnls.2020.3027314
[3] Zhang, Q., Wang, X., Wu, Y. N., Zhou, H., & Zhu, S.-C. (2020). Interpretable CNNs for Object Classification. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1–18. https://doi.org/10.1109/tpami.2020.2982882
[4] Zeldes, Y. (2018). Using Uncertainty to Interpret your Model. KDnuggets. https://www.kdnuggets.com/2018/11/using-uncertainty-interpret-model.html
[5] Johnson, J. (2020). Interpretability vs Explainability: The Black Box of Machine Learning. BMC Blogs. https://www.bmc.com/blogs/machine-learning-interpretability-vs-explainability/#:%7E:text=Interpretability%20vs%20Explainability%3A%20The%20Black%20Box%20of%20Machine%20Learning,-July%2016%2C%202020&text=Interpretability%20has%20to%20do%20with,Nets%2C%20to%20justify%20the%20results.
[6] What is explainable AI (XAI) — and why is it more necessary than ever? (2019, September 5). BBVA |. https://www.bbva.com/ndb/en/article/what-is-explainable-ai-xai-and-why-is-it-more-necessary-than-ever/#:%7E:text=XAI%20is%20a%20great%20tool,a%20new%20world%20of%20possibilities.
댓글