摘要
神经网络作为模式识别、数据挖掘等方面的有效工具,已被广泛应用到生物序列的模式分析中,而生物序列的超大规模、超长同时也给神经网络提出了挑战,即必须解决训练时间过长、效率低下的问题。本文提出了若干适合生物应用的神经网络并行训练策略,并按其神经网络粒度进行分类,同时分析和比较了各种策略的代价。
As an analysis method , neural network has been successfully used in the field of bioinformatics, such as recognition of gene and promoter in DNA sequence, classification of DNA and protein sequences. In the field the bio-sequences are very long and in very large scale, some even up to 6Gbps , and with the rapid development of sequencing technology of the genes, a huge amount of public bio-sequence data are available. The extra huge scale and extra long size of bio-sequences provide some challenges on the neural network. Thus it is very important to reduce the network training time to meet the requirements of bioinformatics and parallel neural network seems one of promising approaches to the problem. In this paper, we have summarized some parallel training strategies of neural network suitable for bioinformatic applications. We also classified them in terms of the granularity of the neural network, analyzed and compared the cost of each strategy.
出处
《计算机科学》
CSCD
北大核心
2004年第3期130-133,178,共5页
Computer Science
基金
国家自然科学基金(60273079)
