Research Associates: H. Byun, S. Kortesis
Research Assistants: T. Papatheodorou, N. Tsantanis, M. Vavalis
Sponsors: ESPRIT, NATO, PRF
This project is part of a large scale effort known by the acronym GALATEA to develop neural networks and virtual neural net environments to support image processing, speech, and character recognition applications. This group is involved with the development of a virtual machine language to support the processing of neural network computations on a variety of parallel machines. An extensive set of this language is currently implemented on nCube 2, Intel IPC/i860, DELTA, and MasPar machines together with a neural network library which includes the most common networks. The performance evaluation of neural net computations and machines is underway. The problem solving environment of GALATEA will be interfaced with these parallel machines providing a machine independent environment. In addition the basic research is conducted on determining the structure of an object from its projections along straight lines. This problem arises in a variety of applications in computer vision and robotics. We study this problem in the context of automatic placement of chips on computer boards using neural network approaches.