首页> 外文学位 >The hypernetwork architecture: A hierarchical molecular interaction model of biological information processing.
【24h】

The hypernetwork architecture: A hierarchical molecular interaction model of biological information processing.

机译:超网络体系结构:生物信息处理的分层分子相互作用模型。

获取原文
获取原文并翻译 | 示例

摘要

A novel architecture for machine learning, the hypernetwork architecture, has been specially designed and implemented in this study. This is a multi level, vertical model of a biological information processing system that includes the flow of information and feedback regulation control inspired by biological systems. The levels considered are the molecular, cellular and organismic.; The molecular level consists of many molecules, i.e., binary string representations derived from enzyme-like structures. Each molecule has an excitatory and a catalytic site, but each has an optional inhibitory site. A molecular interaction, binary string matching, represents a biomolecular self-assembly process. Dynamic formation of networks of molecular interactions represents reaction cascades in biological cells.; Molecules are placed in cells modeled by cellular automata, and an organized group of cells forms an organism. Cell to cell interactions are produced by effector-receptor molecules of the cells.; External influences on the receptor molecules of input cells dynamically trigger cascades of molecular interactions inside the cells of the organism. Then the cascade activates readout molecules on the output cells to form the output of the cell.; Hypernetwork organisms learn classification tasks by means of a variation-selection algorithm based on molecular evolution. Each iteration consists of an organism being reproduced with random molecular mutation, and the better one being chosen to perform the task. The mutation-buffering capabilities of the hypernetwork allow to search for optimal peaks in the fitness landscape.; The hypernetwork effectively learns classification tasks such as the (4–10)-input parity problem, the tic-tac-toe endgame problem, the two x two bit multiplier truth table, and a type of double spiral data set. Experimental results show that, in the hypernetwork architecture, learning improves when molecules exhibit inhibitory sites. This improvement is the result of molecular inhibition and negative feedback regulation inside the cells.
机译:在这项研究中,专门设计并实现了一种用于机器学习的新颖架构,即超网络架构。这是生物信息处理系统的多层垂直模型,其中包括信息流和受生物系统启发的反馈调节控制。考虑的水平是分子的,细胞的和有机的。分子水平由许多分子组成,即衍生自酶样结构的二进制串表示。每个分子都有一个兴奋性和催化位点,但每个分子都有一个可选的抑制位点。分子相互作用,二进制字符串匹配,表示生物分子自组装过程。分子相互作用网络的动态形成代表生物细胞中的反应级联。将分子置于通过细胞自动机模拟的细胞中,有组织的细胞群形成生物体。细胞之间的相互作用是由细胞的效应子-受体分子产生的。对输入细胞受体分子的外部影响动态触发了生物体细胞内部分子相互作用的级联。然后,级联激活输出单元上的读出分子以形成单元的输出。超网络有机体通过基于分子进化的变异选择算法学习分类任务。每次迭代都由一个具有随机分子突变的生物组成,而选择一个更好的生物来执行任务。超网络的突变缓冲功能允许在健身环境中搜索最佳峰。超网络有效地学习了分类任务,例如(4-10)输入奇偶校验问题,井字游戏残局问题,2 x 2位乘法器真值表和一种双螺旋数据集。实验结果表明,在超网络体系结构中,当分子表现出抑制位点时,学习会改善。这种改善是细胞内分子抑制和负反馈调节的结果。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号