Interpretable Machine Learning VS Machine Learning Interpretability

Transparent Machine Learning Models VS Post-hoc Explainability techniques for Black-box machine learning models

The duality of interpretable machine learning and machine learning interpretability can be correlated to a transparent model and post-hoc explainability. On expounding, a transparent model deals with the problem of designing a transparent machine learning model whereas post-hoc explainability deals with the problem of explaining black-box machine learning models by external eXplainable AI techniques. Hence transparent models are interpretable by design while black-box models need to be interpreted by post-hoc explainability.

First of all, it is imperative to know the two extreme sides of interpretable learning models. On one side, we have opaque learning models where the user is unaware of the functioning inside of the learning system. On the contrary, we have comprehensible learning models where the learning model in addition to the regular learning model’s output provides the rules and mapping between different neurons in case of neural networks so that the user can easily comprehend the inner working of the learning model. Interpretable learning models sit in the middle of the above two fringes where the learning model can be analyzed by the user to understand the underlying mapping of the learning model.

Interpretable Machine Learning models can have different levels of transparency. Three major levels of transparency achieved by an interpretable machine learning model are explained below.

Machine Learning Interpretability is useful when the learning model is not interpretable by design and hence the user has to employ external interpretability techniques. The interpretability technique to be used for a machine learning model’s interpretability is decided based upon 1) the users intention i.e. how the user wants the learning model to be explained (by text or visualizations), 2) the procedure to be used i.e. attention analysis and 3) the type of data accepted by the learning model. Post-hoc explainability techniques are majorly divided into the following.

[1] Arrieta, Alejandro Barredo, et al. “Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI.” Information Fusion 58 (2020): 82–115.

Related posts:

Breastfeeding or first infant formula should be continued until your baby is at least one year old. Breastfeeding will benefit you and your kid for as long as you continue. Your baby’s desire for…

If I were to ask you what the answer to a simple math problem was say, one plus one, and we were to get different answers, presumably one of us is wrong. We’re on a spinning rock in the middle of an…

Beberapa minggu lalu gua sibuk namatin series bacaan gua di platform kepenulisan. Di series itu gua udah mulai prejudis kepada tokoh utama. Iya, gua inget dari chapter awal gua mulai menerka-nerka…