Bayesian Knowledge Base (BKB)

A Bayesian Knowledge Base (BKB) is a highly flexible, intuitive, and mathematically sound representation model of knowledge involving uncertainties. It can be depicted as a directed graph with an "if-then" system for determination which allows for domain incompleteness while retaining consistency throughout the model.

Bayesian Knowledge-Base was formulated and developed by Drs. Eugene Santos Jr. and Eugene S. Santos, in 1998, as a powerful alternative to Bayesian Network. Since most knowledge-bases are inherently incomplete, a unique advantage of Bayesian Knowledge Bases (BKBs), in contrast to Bayesian Networks (BNs), is that BKBs do not require a complete specification of the probability distribution for all the random variables (rvs) involved. In other words, BNs cannot be used to deal with incomplete knowledge while BKBs can. Moreover, since only acyclic knowledge is allowed in BNs, while cyclicity of knowledge has a widespread occurrence in most real-world knowledge bases, many common situations are out of reach for BNs, unless it went through additional and unnatural operations. On the other hand, BKB's reasoning scheme is based on inference graphs, which is capable of handling cyclicity. Recently, it has also been shown that unlike BNs, several different BKBs can be readily fused into a single BKB.

As stated above, a BKB represents knowledge in an "if-then" fashion, and may be viewed as a set of conditional probability rules (CPRs) linking the rvs. BKBs can be graphically represented as directed graphs with two primary types of nodes: (1) Instantiation-notes (I-Nodes), which represent the various states of rvs and (2) Support-nodes (S-Nodes), which represent probability values in CPRs.

BKBs have been successfully applied in many important domains, including but not limited to Modeling Care Professional Beliefs and Biases and Malpractice in Surgical settings, 2006 War in Somalia, Modeling and Simulating Behavior of (IL-) legal cross-border migration during 2009 H1N1 Pandemic on US-Mexico border, 2007 South Carolina Democratic Presidential Primaries, Tool to Support Commander's Predictive Environment, Modeling Showdown with Iran over Strait of Hormuz, Embedded Gaming Simulation of Multiple Forces in mid-2000 Baghdad-Civilization 4-based scenario simulation for US-Allied Forces, Support for Effects-Based Operations through Adversarial Intent, Gap Analysis for Multi-Surgeon Teams, Reducing Surgical error through Teams Intent Gap Analysis, Modeling Human Error in Surgical Teams, etc.

BKBs have also been applied widely to construct knowledge bases containing uncertain and/or incomplete knowledge, when other formulations fail. Moreover, by endowing the I-nodes and S-nodes in a BKB with different semantics, BKBs can be used to provide new and distinctive approaches and solutions to complex and diverse problems with significantly superior results.