Inspired by immunity
Inspired by immunity
复制标题
DOI：
10.1038/415468a
复制
发表时间：
2002-01
期刊：
影响因子：
64.8
通讯作者：
E. Klarreich
中科院分区：
文献类型：
--
作者：
E. Klarreich
© 2002麦克米伦杂志有限公司正常的行为，但捕捉欺诈性交易，系统开始把它的规则的组成部分随机，就像免疫系统创建受体或抗体随机基因洗牌。为了淘汰受正常交易刺激的检测器，系统然后将新的检测器发送到已知合法交易的数据库中“存活”一段时间;任何识别出其中一个的检测器都会被丢弃。一旦识别正常事务的检测器被移除，剩余的检测器就会被发送出去监视系统。就像淋巴细胞不受抗原刺激会在几天内死亡一样，没有遇到匹配交易的检测器最终会被杀死，为似乎更接近实际出现的欺诈类型的规则让路。在一个类似于克隆选择和抗体反应改进的过程中，被异常交易警告的检测器会复制自己，添加微小的随机突变，这可能会提高它们的检测能力。成功的检测器作为“记忆细胞”保留在系统中，因此如果再次检测到相同类型的欺诈，系统将迅速做出反应。通过这种方式，它学会识别常见的欺诈模式。例如，如果某个欺诈者通常在深夜进行交易，系统将进化出大量受深夜交易刺激的检测器，大多数免疫启发的安全系统在基本轮廓上是相似的，但都包含了免疫系统的微妙不同属性。伦敦国王学院的理查德·奥弗里尔说：“我们这里有太多美妙的东西了。”Overill的系统，他的团队本月开始与Consignia进行测试，专注于克隆选择和免疫记忆。“人工免疫系统有潜力变得非常聪明和灵活，”他说。其他研究人员关注的不是免疫系统识别异常行为的能力，而是它在没有中央组织机构的情况下协调大量自主细胞产生复杂反应的技能。免疫系统主要通过局部相互作用和化学信号来做到这一点，主要由其细胞分泌的蛋白质介导。“免疫系统没有老板，”以色列Rehobaly魏茨曼科学研究所的免疫反应模型Lee Segel说。“它有一万亿个细胞到处跑，没有人告诉他们该做什么，但仍然做得很好。”
© 2002 Macmillan Magazines Ltd normal behaviour but catch fraudulent transactions, the system starts by putting together the components of its rules randomly, just as the immune system creates receptors or antibodies by random genetic shuffling. To weed out detectors that are stimulated by normal transactions, the system then sends the new detectors to ‘live’for a period of time in a database of transactions that are known to be legitimate; any detector that recognizes one of these is thrown away. Once the detectors that recognize normal transactions are removed, the remaining detectors are sent out to monitor the system. Just as lymphocytes that are not stimulated by an antigen die within days, detectors that do not encounter matching transactions are eventually killed off, to make way for rules that seem to match more closely the types of fraud that actually arise. In a process that is analogous to clonal selection and the refinement of antibody responses, detectors that are alerted by unusual transactions make copies of themselves, adding tiny random mutations that could potentially improve their detective ability. Successful detectors remain in the system as ‘memory cells’, so that if the same kind of fraud is detected again, the system will respond swiftly. In this way, it learns to recognize common patterns of fraud. If, for example, a certain fraudster typically makes transactions late at night, the system will evolve a large number of detectors that are stimulated by late-night transactions.Most immune-inspired security systems are similar in basic outline, but all incorporate subtly different attributes of the immune system.“There are just so many wonderful things we’ve got here,” says Richard Overill of King’s College London, who leads a team that is working to detect fraudulent post-office transactions. Overill’s system, which his team began testing with Consignia this month, concentrates on clonal selection and immunological memory.“Artificial immune systems have the potential to be extremely smart and flexible,” he says. Other researchers are focusing not on the immune system’s ability to identify abnormal behaviour, but on its skill in coordinating huge numbers of autonomous cells to create complex responses without a central organizing body. The immune system does this chiefly through local interactions and chemical signals, largely mediated by proteins secreted from its cells.“The immune system has no boss,” says Lee Segel, who models immune responses at the Weizmann Institute of Science in Rehovot, Israel.“It has a trillion cells all running around, with no one telling them what to do, but still doing a pretty good job.”