M.C.J.D. van Eekelen

CLEAN: A language for functional graph rewriting

Clean is an experimental language for specifying functional computations in terms of graph rewriting. It is based on an extension of Term Rewriting Systems (TRS) in which the terms are replaced by graphs. Such a Graph Rewriting System (GRS) consists of a, possibly cyclic, directed graph, called the data graph and graph rewrite rules which specify how this data graph may be rewritten. Clean is designed to provide a firm base for functional programming. In particular, Clean is suitable as an intermediate language between functional languages and (parallel) target machine architectures. A sequential implementation of Clean on a conventional machine is described and its performance is compared with other systems. The results show that Clean can be efficiently implemented.

The Dutch parallel reduction machine project

In November 1984 three research groups at the universities of Amsterdam, Nijmegen and Utrecht started a cooperative project sponsored by the Dutch Ministry of Science and Education (Science Council). The first phase lasting until the end of 1987 is a pilot study and has as aim to answer the following question. Is it possible and realistic to construct an efficient parallel reduction machine? The present paper gives an outline of the problems concerning parallel reduction machines and of our research towards their solutions.

LEAN: An intermediate language based on graph rewriting

Abstract LEAN is an exaerimental language for specifying computations in terms of graph rewriting. It is based on an alternative to Term Rewriting Systems (TRS) in which the terms are replaced by graphs. Such a Graph Rewriting System (GRS) consists of a set of graph rewrite rules which specify how a graph may be rewritten. Besides supporting functional programming, LEAN also describes imperative constructs and allows the manipulation of cyclic graphs. Programs may exhibit non-determinism as well as parallelism. In particular, LEAN can serve as an intermediate language between declarative languages and machine architectures, both sequential and parallel. This paper is a revised version of a paper by the same authors which was presented at the ESPRIT, PARLE, Conference in Eindhoven, The Netherlands, June 1987.

Operational machine specification in a functional programming language

This paper advocates the use functional programming languages for the formal specification of (abstract) machines. The presented description method describes machines by a two‐level model. At the bottom layer machine components and the micro instructions to handle them are described by using an abstract data type. The top layer describes the machine instructions in terms of these micro instructions.

Functional Descriptions of Neural Networks

In this paper the use of functional programming languages is proposed for the formal specification of neural networks. A formal specification written in a functional programming language has many advantages. First of all, a very high-level, compact, mathematically based description of the functional behaviour of a network and its components can be given. This is due to the presence of higher-order functions and the possibility to define high-level tools to manipulate functions. The compactness is illustrated by the fact that this paper contains a complete description of an xor network. Another advantage of such a specification is that it is directly executable. Hence it can serve as a prototype implementation of the network. Partial correctness of the specification (such as type consistency) is checked automatically by the compiler of the functional programming language. A specification in a functional language inherently makes it possible to combine classical computations with neural computations.

Towards Practical, Precise and Parametric Energy Analysis of IT Controlled Systems

Energy consumption analysis of IT-controlled systems can play a major role in minimising the overall energy consumption of such IT systems, during the development phase, or for optimisation in the field. Recently, a precise energy analysis was developed, with the property of being parametric in the hardware. In principle, this creates the opportunity to analyse which is the best software implementation for given hardware, or the other way around: choose the best hardware for a given algorithm. The precise analysis was introduced for a very limited language: ECA. In this paper, several important steps are taken towards practical energy analysis. The ECA language is extended with common programming language features. The application domain is further explored, and threats to the validity are identified and discussed. Altogether, this constitutes an important step towards analysing energy consumption of IT-controlled systems in practice.

Transforming neural network specifications to parallel programs

Abstract In this paper, we introduce program transformations as a method to turn high level functional specifications of neural networks into efficient sequential and parallel programs. Due to the semantics preserving properties of the applied program transformations, the obtained functional programs have the same meaning as the specifications. This sharply contrasts with the very tedious task of proving imperative programs correct. Program transformations also provide a flexible control over the obtained parallelism. They provide an easy way to experiment with various distributions of neural networks on parallel computer architectures.