A systolic algorithm for back-propagation: mapping onto a transputer network
This paper is devoted to the implementation of back-propagation - a supervised learning algorithm for multilayered feed-forward connectionist networks on local memory multiprocessor systems. First, a systolic algorithm is described, where dependencies are considered at the data item level. The systolic array is then partitioned and mapped onto a multiprocessor system. At this stage, the level of granularity is increased, in order to reduce communication cost. Finally, each stage is implemented on a transputer-based multiprocessor, and their performance is compared with a simple sequential version of the learning rule. A parallelisation rate of about 0.9 is obtained.
Bibliographic Reference: Paper presented: Joint IEE and OUG Colloquium, London (GB), Oct. 1, 1990
Availability: Available from (1) as Paper EN 35813 ORA
Record Number: 199011793 / Last updated on: 1994-12-02
Original language: en
Available languages: en