|
 N° 12 - Décembre 2020 - Intelligences artificielles et humaines, quelles interactions ?
A Note on the Interpretability of Machine Learning Algorithms
Dominique GUÉGAN
To analyze the concept of interpretability associated with an ML algorithm, a distinction is made between “how” (How does a black box or a very complex algorithm work?) and “why” (Why does an algorithm produce such-and-such a result?). These questions appeal to many actors: users, professionals, regulators, etc. Using a formal, standardized framework, existing solutions are indicated by specifying which elements in the supply chain are impacted when answers are provided to the previous questions. This presentation, by standardizing notations, allows for a comparison of different approaches so as to highlight the specificity of each (their objectives and processes). This study is not exhaustive — the subject is far from closed…
 Télécharger gratuitement l'article
 Retour au sommaire
 N° 12 - December - Artificial and human intelligences: What interactions?
Note on the interpretability of machine-learning algorithms
Dominique GUÉGAN
To analyze the concept of interpretability associated with an ML algorithm, a distinction is made between “how” (How does a black box or a very complex algorithm work?) and “why” (Why does an algorithm produce such-and-such a result?). These questions appeal to many actors: users, professionals, regulators, etc. Using a formal, standardized framework, existing solutions are indicated by specifying which elements in the supply chain are impacted when answers are provided to the previous questions. This presentation, by standardizing notations, allows for a comparison of different approaches so as to highlight the specificity of each (their objectives and processes). This study is not exhaustive — the subject is far from closed…
Download full article
Retour au sommaire
|
