Alapfogalmak
The EDGE language provides a unified, mathematical framework for expressing graph algorithms using extended Einsum notation from tensor algebra, enabling a separation of concerns between what to compute and how to compute it.
Kivonat
The paper proposes the EDGE language, which extends Einsum notation from tensor algebra to provide a unified, mathematical framework for expressing graph algorithms. The key ideas are:
Leveraging the graph-matrix duality, where a graph can be represented as a 2D tensor.
Extending the expressive power of Einsum notation to support more complex operations common in graph algorithms.
The EDGE language aims to:
Allow researchers to more easily compare different graph algorithms and implementations.
Enable developers to separate the concerns of what to compute (described with EDGE notation) from how to compute it.
Enable the discovery of different algorithmic variants of a problem through algebraic manipulations and transformations on EDGE expressions.
The paper first provides background on graphs, tensors, and the tensor algebra space. It then discusses the limitations of current Einsum notation in expressing graph algorithms and outlines the design goals for the EDGE language. The key extensions to Einsum notation in EDGE include:
Support for user-defined data values and types
Initialization of tensors
Generic user-defined operators for mapping, reduction, and assignment
Separation of computations on graph entities/tensor coordinates from graph values/tensor data
Iterative algorithms
Constraints based on rank variable expressions
The authors envision an end-to-end EDGE framework that leverages the separation of concerns to enable exploration of the algorithmic and implementation space through algebraic manipulations and transformations of EDGE expressions.
Statisztikák
There are no key metrics or figures used to support the author's main arguments.
Idézetek
There are no striking quotes supporting the author's key logics.