A Multi-Answer Character Recognition Method and Its Implementation on a High-Performance Computing Cluster
Qing Wu, Morgan A. Bishop, Robinson E. Pino, Richard Linderman, Qinru Qiu · 2011
Abstract — In this paper, we present our work in the implementation and performance optimization of a novel multi-answer character recognition method on a high-performance computing cluster. The main algorithm used in this method is called the Brain-State-in-a-Box (BSB), which is an auto-associative neural network model. We applied optimization techniques on different parts of the BSB algorithm to improve the overall computing and communication performance of the system. Furthermore, the proposed method adopts a new way to train, recall, and organize the BSB models for different characters, in order to provide a sorted (based on recall convergence speed) list of candidates for a given character image.