项目名称: 矩阵补全中的非凸、随机和在线方法
项目编号: No.61472347
项目类型: 面上项目
立项/批准年度: 2015
项目学科: 自动化技术、计算机技术
项目作者: 钱徽
作者单位: 浙江大学
项目金额: 80万元
中文摘要: 矩阵补全问题研究具有理论意义和应用价值。目前,随着应用领域对求解精确性、稳定性和可扩展性的要求不断提高,针对矩阵补全问题的研究出现了新的发展方向。从问题模型角度观察,非凸惩罚模型为问题求解的精确性和稳定性提供了可能性,正成为矩阵补全中的一个热点研究方向。从计算方法的角度看,随机矩阵补全方法,在分布式计算架构的支持下,提供了解决大规模数据集恢复问题的可能性,正在形成矩阵补全研究中的一个重要研究方向。此外,许多应用呈现出的增量和流式数据输入的特点导致了在线矩阵补全方法的发展,成为领域专家关注的新研究方向。因此,本研究关注低秩矩阵补全中的非凸、随机和在线方法。研究将重点关注矩阵补全方法中的非凸惩罚技术、层次贝叶斯模型、随机矩阵分解和非凸在线方法。这些方法将丰富矩阵补全技术领域的研究成果,同时为应用提供支持。
中文关键词: 低秩;矩阵补全;非凸;随机算法;在线学习
英文摘要: Matrix completion is an important theme both theoretically and practically. However, the state-of-the-art methods based on convex optimization usually can neither lead to a unbiased solution nor handle large-scale dataset. Moreover, seldom current algorithms can fulfill the requirement from online operation due to their batch paradigm. For this reason, this project intends to explore the problem of finding both non-convex alternative to approximate the matrix rank and stochastic methods for real-world application with massive data flow. And motivated by the recent developments of online learning theory, we also have interests in the incremental matrix completion. We will concentrate on non-convex penalty theory, Hierarchical bayes, and stochastic matrix factorization and online learning for non-convex optimization. Robust methods will be proposed with rigorous theoretical proof. The proposed methods are expected to enrich the theoretical achievement in the field of Matrix completion, and to provide technical support for the applications such as social networking, intelligent robotics, and image processing.
英文关键词: Low-rankness;Matrix Completion;Non-convexity;Stochastic Algorithm;Online Learning