Abstract

I plan to talk about how the underlying types can be reconstructed using a probabilistic model based on the values we find in a program or a database. Furthermore, in the case where no types existed beforehand, we can construct types based on the values we find.

This is useful for databases, for example, in constructing a more normalised database given some partially filled in tables. It maps particularly well onto key-value pair stores. We also envisage it could be used in type reconstruction in scripting languages.

The talk will look at the motivations for doing this, the probabilistic technique used, and talk about some of the further work planned. I hope it will not be too long, and there will be time to get people’s input on how this might be used, whether anything like this has been done before, and how it could be expanded to cover more complex type systems.