Explainable Artificial Intelligence in Industry
Open Track - ISM International Conference of Smart Manufacturing
http://www.msc-les.org/ism2020/ - Contact: florian.sobieczky@scch.at

Explainable artificial has emerged as a key subject for all fields in which the value of accuracy of high performing predictive machine learning tools (such as deep learning) is compromised by a lack of the performance’s interpretability. While due to legal implications the hype of XAI has conquered mobile communication technology, medical research, and autonomous driving, it has now also reached operations research.
In manufacturing, there exists a very high demand for explanations of the machine learning predictions concerning waste, mal-performance of machinery, and need for maintenance, as they can be used in a more strategic way as opposed to only on the process control level. If the (human) operators are learning from the black box –instead of merely receiving corrections– they can remove the causes for the diagnosed errors. This means using artificial intelligence in an effective, sustainable way. The session ‘Explainable Artificial Intelligence in Industry’ focusses on insight into the methods and procedures illuminating the causes for Machine Learning algorithms’ decisions.
Scope:
1. XAI and interpretable machine learning models in manufacturing
2. Criteria for explainability (Interpretability, trustworthiness, transparency, faithfulness, stability, counter-factual)
3. Local surrogate models: Linearity, Fidelity, Stability
4. Model-agnosticity
5. Probabilistic Methods: Partial dependence plots, individual conditional expectation, accumulated local effects plots
6. Game Theoretic Methods: Shapley values
7. Robustness of Interpretability
8. Salience maps, interpretability in vision
9. Human in the loop, Interpretability and Usability