- The Bayesia Product Portfolio
- BayesiaLab Knowledge Hub & Library
- BayesiaLab Community
- Courses & Events
- BayesiaLab Store
Bayesian Expert System for Troubleshooting
The BEST suite consists of a set of software modules for the design, administration and operational deployment of intelligent troubleshooting applications for complex technical domains. It provides powerful and easy-to-use tools for modeling and disseminating expert knowledge. BEST features machine learning capabilities that can automatically improve models using in-service experience. Additionally, BEST can generate troubleshooting strategies that minimize the cost of performing diagnosis.
Invented by Judea Pearl in the 1980s at UCLA, Bayesian networks are a mathematical formalism that can simultaneously represent a multitude of probabilistic relationships between variables in a system. The graph of a Bayesian network contains nodes (representing variables) and directed arcs that link the nodes. The arcs represent the relationships of the nodes.
Whereas traditional statistical models are of the form y=f(x), Bayesian networks do not have to distinguish between independent and dependent variables. Rather, a Bayesian network approximates the entire joint probability distribution of the system under study.
Researchers can use BayesiaLab to encode their domain knowledge into a Bayesian network. Alternatively, BayesiaLab can machine-learn a network structure purely from data collected from the problem domain. Irrespective of the source, a Bayesian network becomes a representation of the underlying, often high-dimensional problem domain. The researcher can then use BayesiaLab to carry out “omni-directional inference,” i.e., reason from cause to effect (simulation), or from effect to cause (diagnosis), within the Bayesian network model.
On this basis, BayesiaLab offers an extensive analytics, simulation and optimization toolset, providing comprehensive support for policy development and decision making. In this context, BayesiaLab is unique in its ability to distinguish between observational and causal inference. Thus, decision makers can correctly simulate the consequences of actions not yet taken.