9560789 Buntine This Small Business Innovation research Phase I project is to develop efficient, effective applications of machine learning which is an expensive, error-prone process even for highly-trained experts. It has been previously shown that probability networks can represent a broad variety of learning and discovery problems, and manually transforming a network can yield algorithms for recovering or discovering model parameters from data. This new approach has been demonstrated in basic research, but not yet generalized and automated in order to test applicability and feasibility. This effort will complete the research and development to turn these results into a commercial application: an automated CASE-tool for researchers and developers in machine learning and knowledge discovery. The design goal is a Probability Network Language (PNL), a front-end graphical prototyping tool, and a PNL compiler that emits efficient C or C++ code for learning and discovery applications. If the initial effort is successful, The follow on effort will produce end-to-end functionality in a complete prototype for PNL. PNL is precisely what is needed to make machine learning and knowledge discovery a practical technology, accessible to a wide range of application programmers. PNL promises significant benefits to engineers and scientists in the U.S., particularly those involved in data analysis, but also those faced with complex diagnostic problems, real-time control problems, and voluminous or heterogeneous data. If it proves to be a viable and effective tool for rapid prototyping of learning and discovery applications, PNL will provide the missing tools needed for efficient development of state-of-the-art applications in diagnosis control and data analysis. It is intended to use and license PNL in commercial applications for scientific research, financial services and biotechnology.
