5
$\begingroup$

Explainable artificial intelligence (XAI) is concerned with the development of techniques that can enhance the interpretability, accountability, and transparency of artificial intelligence and, in particular, machine learning algorithms and models, especially black-box ones, such as artificial neural networks, so that these can also be adopted in areas, like healthcare, where the interpretability and understanding of the results (e.g. classifications) are required.

Which XAI techniques are there?

$\endgroup$
11
$\begingroup$

Explainable AI and model interpretability are hyper-active and hyper-hot areas of current research (think of holy grail, or something), which have been brought forward lately not least due to the (often tremendous) success of deep learning models in various tasks, plus the necessity of algorithmic fairness & accountability.

Here are some state of the art algorithms and approaches, together with implementations and frameworks.


Model-agnostic approaches

SHAP seems to enjoy high popularity among practitioners; the method has firm theoretical foundations on co-operational game theory (Shapley values), and it has in a great degree integrated the LIME approach under a common framework. Although model-agnostic, specific & efficient implementations are available for neural networks (DeepExplainer) and tree ensembles (TreeExplainer, paper).


Neural network approaches (mostly, but not exclusively, for computer vision models)


Libraries & frameworks

As interpretability moves toward the mainstream, there are already frameworks and toolboxes that incorporate more than one of the algorithms and techniques mentioned and linked above; here is a partial list:


Reviews & general papers


eBooks (available online)


Online courses & tutorials


Other resources

| improve this answer | |
$\endgroup$
3
$\begingroup$

There are a few XAI techniques that are (partially) agnostic to the model to be interpreted

There are also ML models that are not considered black boxes and that are thus more interpretable than black boxes, such as

  • linear models (e.g. linear regression)
  • decision trees
  • naive Bayes (and, in general, Bayesian networks)

For a more complete list of such techniques and models, have a look at the online book Interpretable Machine Learning: A Guide for Making Black Box Models Explainable, by Christoph Molnar, which attempts to categorise and present the main XAI techniques.

| improve this answer | |
$\endgroup$
0
$\begingroup$

Jacques Pitrat's last book Artificial Beings The Conscience of a Conscious Machine has some interesting chapters (§2, §3, §5, §6, §7) related to the questions of explanations and meta-explanations in AI systems.

It describes a reflexive, "expert system" like, meta-knowledge based approach to explainable AI

You could also read his papers Implementation of a reflective system (1996) and A Step toward an Artificial AI Scientist online (there could be a typo in it: "pile" is the French word for stack, including the call stack).

You might also look into J.Pitrat's blog and into the ongoing RefPerSys project (it is in November 2020).

PS. J.Pitrat (born in 1934) passed away in October 2019. French readers could see this. His blog might disappear in a few months.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.