Explainable AI: Foundations, Methodologies and Applications by Unknown

Author:Unknown
Language: eng
Format: epub
ISBN: 9783031128073
Publisher: Springer International Publishing

6.3.2 Explainability by Using White-Box Models

White-box models are transparent models and are easily understandable or interpretable. This category contains mainly linear models, decision trees, and some complex models that are easy to interpret. Some of these complex models used in medical imaging and healthcare are listed here.

Microsoft came up with an interpretable model for predicting pneumonia risk, which also had great accuracy (Caruana et al. 2015). The authors discussed that while interpretable linear models and decision trees could not give good results, neural nets gave much better results, but at the cost of explainability. High-performance generalized additive models with pairwise interactions (GA2Ms) were proposed and tested on two real medical data case studies. The authors also mentioned that the model could be scaled to work on thousands of patient data without losing accuracy and still being highly interpretable.

Another technique that utilizes Boolean rules to create predictive models was proposed—Boolean Rule Column Generation (Dash et al. 2018). This technique uses easy-to-understand Boolean rules with some clauses and conditions. Humans easily understand these clauses and conditions. GLM or Generalized Linear Rule Models (Wei et al. 2019) use an ensemble of rule-based features. GLMs are easy to interpret and complex simultaneously, as the rules can capture non-linear dependencies. TED or Teaching Explanations for Decisions (Hind et al. 2019) is a framework that tries to produce explanations like a human expert rather than explaining the inner workings of an AI model.

Not much research has been done in the complex white-box model development domain. No white-box model can produce the same high accuracy as deep learning models. The white-box models are also very domain-specific, unlike various computer vision and natural language processing neural networks used on various real-world tasks.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.