Graphical Framework to Incorporate Knowledge Sources
In this chapter, we introduce the design of our proposed framework, the so called GFIKS (graphical framework to incorporate additional knowledge sources). It is based on a graphical model representation that makes use of additional knowledge sources in a statistical model as shown in Figure 3.1. This approach is meant to be broadly useful in the sense that it can be applied to many existing modeling problems with their respective model-based likelihood functions.
In Section 3.1, we review graphical model representation, including probability theory and graph theory. In Section 3.2, we introduce GFIKS’s procedure for knowledge incorporation, including how to dene the causal relationships between information sources (Section 3.2.1), how to do direct inference (Section 3.2.2), and how to proceed when direct inference is intractable (Sections 3.2.3 and 3.2.4). In Section 3.3, we discuss the general issues and possibilities of incorporating knowledge sources in the statistical ASR system, such as what type of knowledge sources are important (Section 3.3.1) and at which level of ASR they should be incorporated (Section 3.3.2)
KeywordsBayesian Network Knowledge Source Graphical Transformation Direct Inference Junction Tree
Unable to display preview. Download preview PDF.