Change actions: from incremental computation to discrete derivatives
share
The goal of this thesis is threefold: first, to provide a general semantic setting for reasoning about incremental computation. Second, to establish and clarify the connection between derivatives in the incremental sense and derivatives in the analytic sense, that is to say, to provide a common definition of derivative of which the previous two are particular instances. Third, to give a theoretically sound calculus for this general setting. To this end we define and explore the notions of change actions and differential maps between change actions and show how these notions relate to incremental computation through the concrete example of the semi-naive evaluation of Datalog queries. We also introduce the notion of a change action model as a setting for higher-order differentiation, and exhibit some interesting examples. Finally, we show how Cartesian difference categories, a family of particularly well-behaved change action models, generalise Cartesian differential categories and give rise to a calculus in the spirit of Ehrhard and Regnier's differential lambda-calculus.
READ FULL TEXT