Networks are one of the most intuitive representations of complex data. However, most networks rely on pair-wise associations which limits their use for making decisions. Bayesian Decision Networks (BDNs) extend a class of probabilistic graphical models known as Bayesian Networks by using decision theory and have been used in business settings for decision making. However, BDNs are under-exploited in clinical and public health settings because of the complex nature of datasets which makes it difficult for these networks to be hand-specified. This tutorial will teach the participants to learn Bayesian-network models directly from data, assess these rigorously with statistical bootstrap evaluations, draw quantitative inferences, learn optimal decisions and deploy their models as a web-application based upon R/Shiny framework. Participants will learn to use these models both for probabilistic reasoning and causal inference depending upon the study design. Since a BDN is a directed acyclic graph and provides a single joint-multivariate fit on the data, it automatically learns the confounder, mediator and collider effects in the data, hence providing an end-to-end statistical and machine learning framework for knowledge-discovery in addition to decision making. Fitting a joint probabilistic model decreases the chance of false edges because the structure has to agree with global and local distributions. Unlike most other forms of Artificial Intelligence and Machine Learning, BDNs are white-box models falling in the class of Explainable AI (XAI) and Fair Accountable Transparent ML (FAT-ML). The tutorial will cover an end-to-end walkthrough of the open-source platform, wiseR (Figure. 1) developed by the instructor and his team in collaboration with computer scientists and clinicians at Stanford and India. The tutorial will cover preliminary theory and two case-studies, in a clinical setting for Sepsis and a public health setting (Health Inequality) for learning decisions and policy, both published and available with linked open-data.

Describe the new knowledge and additional skills the participant will gain after attending your presentation.: Participants will be able to understand and use Bayesian Decision Networks in the clinical settings for:-
1. Probabilistic reasoning with complex datasets.
2. Causal reasoning in complex datasets with interventions
3. Decision making under uncertainty
4. Deploying their models as Bayesian AI enabled dashboards as web-services.


Tavpritesh Sethi (Presenter)
Stanford University

Presentation Materials: