分布式并行计算领域存在着需要在可承受的时间范围内，求解出超大规模数据集(样本数量达到几十亿的规模级别或更多)的逻辑回归模型的需求。本项目致力于用交替方向乘子法框架设计近似求解L2范数正则化逻辑回归算法来满足这种新兴的需求，其研究内容主要为算法设计，分为如下三部分：1.如何将求解原问题等价地转换为求解一系列子问题的目标能量函数，这决定了算法的根本性框架；2. 如何高效地近似求解每一种类别的子问题；3. 研究设计的算法的各方面性能，包括误差的阶、精确度、模型泛化能力、迭代收敛性和收敛率和时间复杂度等。本研究是连接凸优化逻辑回归算法设计领域和分布式并行计算领域的一座桥梁。.应用逻辑回归求解算法去解决现实工程问题的文献数量可以说是非常巨大的，然而能提出新的求解算法的文献数量却非常有限，而新算法在理论和实际性能上具有实质性突破的研究工作更是凤毛麟角。本项目正是这样的研究工作之一。

In the field of distributed parallel computing, there is a need to solve the logistic regression model for large-scale data sets (the scale of the number of samples reaches billions or more) in an affordable time range. This project is dedicated to design approximate algorithms for solving the L2-norm regularized logistic regression algorithm with the framework of Alternating Direction Method of Multipliers to meet this emerging demand. The research content is mainly about algorithm design, which is divided into three parts as follows: 1. How to transform the original problem of the objective energy function to a series of sub-problems equivalently, which determines the fundamental framework of the algorithm designed; 2 How to solve each kind of sub-problem efficiently; 3. Study the performance of the algorithm, including the order of error, accuracy, model generalization ability, iterative convergence, convergence rate and time complexity. This research is a bridge between the design of convex optimization logistic regression algorithm and the field of distributed parallel computing...The number of papers that employ logistic regression to solve practical engineering problems is very large, but the number of papers that can propose new algorithms is very limited, and the research work that the new algorithm proposed has substantial breakthrough in terms of both theoretical and practical performances is even rarer. This project is just one of such research work.