Giving sense to old databases
with Machine Learning
The goal is to find a tool, which, using AI, walks on the Database and, by applying algorithms, extracts the meaning, including logical meaning, and semantics of the tables contained on the same DB
Dinosaur databases are running the world! As relics of the early steps of the information era, these Databases are still the basis of many economic transactions. Although their age, it seems to be extremely difficult to replace them with novel and faster solutions. These databases have been written in a wonderful era in which memory was a problem. Hence, variable names, table names, and column names were short and cryptic. Moreover, documents describing these names are buried in forgotten places, if still existent. The challenge is, then giving sense to these dinosaur databases, to help software engineers to rediscover the sense of these databases to produce the novel version
Which data we offer you?
We start from an Oracle database resulting from a Db2 import with no indexes, no relations among tables, no integrity checks. Basically, each single table is isolated, without any declared relation with other tables. Additionally, table names and field names are not meaningful at all. Tables are more than a thousand but a potential kernel of 400 seems to be really used. Database is completely undocumented. Application layer accessing this database is written in Cobol and it’s completely undocumented too, but fully available for study
The goal of the research program?
A solution capable of analyzing tables’ structure and, applying a set of inferences or novel algorithms, capable of proposing a potential graph of internal relations among tables. For each table, proposed solution should be able to offer a few potential dependencies towards other tables. More specifically, table fields must be involved. Semantic meaning shoud be guessed for tabels and fields. Final database understanding must be completed by a machine to human supervision process. A human reviser will manually check and validate each proposed link between table. At this level, it’s important that the proposed solution offers an efficient way to inspect tables and their data.
awarded a prize of 50,000 € 10,000 unrestrictited gift for the winning idea
40,000 upon delivery of a prototype
For any question you can email us at email@example.com or attend to one of the
two Q&A video sessions offered on the dates below:
October 10th 2022
November 14th 2022
The participation link will be available here a few days before the event.