Neural Network Processing Using Database Management Systems

Database Management Systems (DBMS) have become an integral part of any data storage, processing, or retrieval system. They are uniquely suited to manipulate large amounts of data while maintaining a level of data integrity and security. Meanwhile, Neural Networks are being used to perform operations on data to discover the unknown underlying input/output relationships. A significant part of Neural Network processing is reducing the input space to a manageable size. These algorithms utilize some of the same techniques that databases use (such as hashing functions) to retrieve or store large amounts of data quickly. This project utilizes a popular DBMS to build a Neural Network application that resides inside the database. This will yield many benefits that weren’t previously attainable, e.g., the data and the Neural Network are both located in the same physical location. There is no need to export the data from the database, manipulate it into an appropriate format, and then use a separate Neural Network application to process the data. Another benefit is that there is no need for database users to learn external applications to process the data from their database. They simply issue an appropriate SQL statement and the Neural Network will perform the calculations and give a result. If the data is changed, the user simply re-issues the SQL statement to recalculate the outputs of the system. Utilizing this system, database users will be able to perform high level data processing tasks using the tools with which they are familiar, while appreciating significant performance gains over other implementations.