The programming environment consists of:
an object oriented programming language, N
an algorithm library written in N
a C-compatible intermediate language (VML)
an interface to a silicon compiler
Industrial applications cover two areas: optical character recognition (OCR) of printed documents and industrial vision for quality control. The industrial vision work focuses on improving both SMD (surface mounted device) assembly techniques and video-based grading systems for fruit.
Progress has been made in the definition of the virtual machine concept allowing different hardware processing entities to participate together to provide the definition of a generic heterogeneous architecture based on neurocomputing operators as well as other processors. Future work concerning the analysis of parameters about communication will allow the determination of the interconnection medium between all these entities. The analysis of neural algorithms has been completed and a set of compute-bound operations identified which are shared by all neural nets. Very large scale integration (VLSI) design of a neural signal processor that executes these operations has started. The functional block architecture of the system has been fixed at chip and board level. The design of all the key components of the software and silicon compiler workpackages has been completed. The preparatory phase for the industrial vision and optical character reader workpackages has been completed. The most challenging problems have been pinpointed and the first neural net solutions proposed.
The objective of the neurocomputing project is the construction of a general purpose neural computing system for Europe. This comprehensive system will encompass:
general purpose neurocomputer hardware (GPNC), which efficiently supports a wide range of neural networks;
a sophisticated neural programming environment allowing the efficient use of systems comprising the GPNC, domain specific processors and application specific integrated circuits (ASIC), plus conventional parallel computers and workstations;
a silicon compiler for the rapid production of ASICs;
3 industrial applications, in the factory and office, that demonstrate the capabilities of the hardware, software environment and silicon compiler, and neural networks.
The industrial applications cover 2 areas, optical character recognition (OCR) and industrial vision. The main results of the workpackage will be a system able to read printed documents and integrate them into any word processing system. The results of the industrial vision workpackage are intended on the 1 hand to improve surface mounted device (SMD) assembly techniques for future products, and on the other hand to improve fruit video grading systems.
The original objectives have now materialized into the aim of having a first GPNC demonstrator in which specialized neural boards developed by 2 partners will be attached to workstations. The demonstrator has been specified and will serve as a vehicle to test the hardware and software concepts. The virtual machine language (VML), a key component, has been specified and the compiler from the high level language N into VML is under development. Work on the silicon compiler has progressed though several options are still open.
In the application areas significant progress was reported in the optical character recognition (OCR) application in particular. The basic building blocks for the application systems have been identified.
Funding Schemeundefined - undefined
20041 Agrate Brianza Milano
WC1E 6BT London