Phishing is the most well-known act of deceiving the Internet users, in which the ‘perpetrator’ plays a credible entity. This is done by misusing the inadequate protection provided by electronic tools, and by exploiting the ignorance of the user-object, in order to illegally obtain personal data, such as sensitive private information and passwords. This research proposes the online meta-learning firewall to prevent phishing attacks. It is a highly innovative and fully automated active safety tool that uses a long short-term memory meta-learner algorithm. This method can learn to efficiently classify using a small number of samples. At the same time, it can converge with a fairly small number of steps. The proposed system is an improvement on the k-nearest neighbor with self-adjusting memory algorithm, which is inspired by the model of short and long-term memory. The purpose of the system is to understand the nature of an unknown situation and to classify it, based on the most relevant characteristics that come directly from the unknown environment.

网络钓鱼是欺骗互联网用户的最著名的行为，其中“犯罪者”扮演着可信的角色。这是通过滥用电子工具提供的保护不足以及利用用户对象的无知来非法获取诸如敏感私人信息和密码之类的个人数据来实现的。这项研究提出了一种在线元学习防火墙，以防止网络钓鱼攻击。这是一个高度创新的，完全自动化的主动安全工具，它使用了长期的短期记忆元学习器算法。该方法可以学习使用少量样本进行有效分类。同时，它可以以很少的步骤收敛。所提出的系统是对k最近邻的一种改进，具有自调整记忆算法，这是受到短期和长期记忆模型的启发。该系统的目的是基于直接来自未知环境的最相关特征，了解未知情况的性质并对其进行分类。