A programming language for precision--cost tradeoffs

Abstract

Many computational systems need to deal with various forms of imprecision and uncertainty in their data; it is also the case that many systems, especially mobile and distributed systems, must be able to trade off the precision of their data and operations against the cost of performing those operations. Unfortunately, for many applications, trying to make these tradeoffs severely complicates the program, because there does not yet exist a programming model that gives the programmer the ability to easily describe the relevant tradeoffs between precision and cost of operations or to express in an algorithm what tradeoffs are appropriate under what circumstances. This paper lays a solid foundation for exploring such programming models by introducing and analyzing a simple core abstraction on which others can be based. We determine what sorts of strategies are and are not possible within this abstraction, and discuss what specific difficulties must be overcome in future work in order to extend the abstraction to encompass a larger class of programs.