Paul Lukowicz

Experimental evaluation in computer science: a quantitative study

A survey of 400 recent research articles suggests that computer scientists publish relatively few papers with experimentally validated results. The survey includes complete volumes of several refereed computer science journals, a conference, and 50 titles drawn at random from all articles published by ACM in 1993. The journals of Optical Engineering (OE) and Neural Computation (NC) were used for comparison. Of the papers in the random sample that would require experimental validation, 40% have none at all. In journals related to software engineering, this fraction is 50%. In comparison, the fraction of papers lacking quantitative evaluation in OE and NC is only 15% and 12%, respectively. Conversely, the fraction of papers that devote one fifth or more of their space to experimental validation is almost 70% for OE and NC, while it is a mere 30% for the computer science (CS) random sample and 20% for software engineering. The low ratio of validated results appears to be a serious weakness in computer science research. This weakness should be rectified for the long-term health of the field. The fundamental principle of science, the definition almost, is this: the sole test of the validity of any idea is experiment. —Richard P. Feynman. Beware of bugs in the above code; I have only proved it correct, not tried it. —Donald E. Knuth

Design of an opto-electronic VLSI/parallel fibre bus

We present a design for an opto-electronic bus system based on commercial parallel fibre ribbon transmission systems and opto-electronic VLSI smart pixel arrays. We describe the design of the individual components, their integration, the performance limits and potential applications.

A Massively Parallel Implementation of the Full Search Vector Quantization Algorithm

We present a massively parallel version of a full search vector quantization and its application in the development of an audio-visual speech recognition system. The parallel implementation reduced the worst case runtime of (estimated) 80–100 hours on a 10 MFLOP SPARC to less then 2 hours on a 2.4 GFOLP MasPar MP2216. This demonstrates how the use of parallel computers reduces product development time and leads to a more mature design by allowing for more extensive experimentation with different data sets.

The Modula-2* environment for parallel programming

Presents a portable parallel programming environment for Modula-2*, an explicitly parallel machine-independent extension of Modula-2. Modula-2* offers synchronous and asynchronous parallelism, a global single address space, and automatic data and process distribution. The Modula-2* system consists of a compiler, a debugger, a cross-architecture make, graphical X Windows control panel, run-time systems for different machines, and sets of scalable parallel libraries. The existing implementation targets the MasPar MP series of massively parallel processors (SIMD), the KSR-1 parallel computer (MIMD), heterogeneous LANs of workstations (MIMD), and single workstations (SISD). We describe the important components of the Modula-2* environment, and discuss selected implementation issues. We focus on how we achieve a high degree of portability for our system, while at the same time ensuring efficiency.<<ETX>>

Project Triton: towards improved programmability of parallel computers

The main objective of Project Triton is adequate programmability of massively parallel computers. This goal can be achieved by tightly coupling the design of programming languages and parallel hardware.

Human Activity Recognition in Smart Environments (Dagstuhl Seminar 12492)

This report documents the program and the outcomes of Dagstuhl Seminar 12492 Human Activity Recognition in Smart Environments. We established the basis for a scientific community surrounding activity recognition by involving researchers from a broad range of related research fields. 30 academic and industry researchers from US, Europe and Asia participated from diverse fields including pervasive computing, over network analysis and computer vision to human computer interaction. The major results of this Seminar are the creation of a activity recognition repository to share information, code, publications and the start of an activity recognition book aimed to serve as a scientific introduction to the field. In the following, we go into more detail about the structure of the seminar, discuss the major outcomes and give an overview about discussions and talks given during the seminar.

On the feasibility of a scalable opto-electronic CRCW shared memory

We discuss the results of a feasibility study of an opto-electronic shared memory with concurrent read, concurrent write capability. Unlike previous such work we consider a true hardware shared memory rather then a simulation on a tightly, optically connected distributed memory computer. We describe a design that could be implemented using compact integrated semiconductor modules and propose ways to solve two major problems faced by such a device: optical system complexity and parallel word level write consistency. It is shown that, in principle, a memory with GBytes capacity and a latency of less then 1 ns, accessed by up to 10/sup 5/ processors could be feasible. Using devices currently available as laboratory prototypes and taking into account energy and crosstalk considerations a capacity of more then 1 MB and a latency of about 50 ns might be attained for up to 1000 processors.<<ETX>>

Using Reduce for Replica Calculations

The REDUCE package REPPACK for performing replica calculations in the context of neural networks is presented. REPPACK provides the user with several functions needed to calcuxad late partition functions and saddle point equations interactively. The application of REPPACK is demonstrated for the toy example of finitely many patterns (pIN - 0) for the Hopfield model. The Gardner calculation with REPPACK is briefly outlined.