Section: New Results

Incrementality

Participants : Yann Régis-Gianas, Lourdes Del Carmen González Huesca, Thibaut Girka.

An optimisation to perform incremental computations was developed by Lourdes del Carmen González Huesca and Yann Régis-Gianas, providing a mechanism to achieve efficiency. Incrementality as a way to propagate an input change into a corresponding output change is guided by formal change descriptions over terms and dynamic differentiation of functions. The data-changes are represented by displaceable types, a general framework to displace terms directed by types. An extension of the simply-typed lambda-calculus with differentials and partial derivatives offers a language to reason about incrementality. The basic system, $\lambda$-diff, was enriched with expressions for fixed-points and data-types together with their corresponding derivatives to analyse incrementality over them. The above results are reported in the second part of Lourdes González Huesca PhD thesis [16] .

In collaboration with Paolo Giarrusso and Yufei Cai (Univ Marburg, Allemagne), Yann Régis-Gianas developed a new method to incrementalise higher-order programs using formal derivatives and static caching. A paper is in preparation.

In collaboration with David Mentré (Mitsubishi), Thibaut Girka and Yann Régis-Gianas designed and certified a new algorithm for correlating program generation: such a program is used to characterise the differences between two close programs. (Therefore, a correlating program is a good input for an incremental static analyser.) Before their work, only one algorithm existed in the literature and it was unsound. The new algorithm is sound and certified in Coq. This work has been published in the ATVA conference. Thibaut Girka has presented this work [33] at ATVA 2015.

In collaboration with David Mentré (Mitsubishi), Thibaut Girka and Yann Régis-Gianas are developing a theoretical framework to define a notion of differential operational semantics: a general mathematical object to characterise the difference of behavior of two close programs.