Block-iterative algorithms
Block-iterative algorithms
复制标题
DOI：
10.1111/j.1475-3995.2008.00683.x
复制
发表时间：
2009-07-01
影响因子：
3.1
通讯作者：
Byrne, Charles
中科院分区：
文献类型：
--
作者：
Byrne, Charles
最近提出的顺序无约束最小化算法 SUMMA 进行了扩展，为块迭代或部分梯度优化方法的推导提供了一个框架。该块迭代 SUMMA (BI-SUMMA) 包括代数重建技术 (ART) 及其乘法变体 MART 的块迭代版本，并受其启发。 BI-SUMMA 方法旨在提供计算上易于处理且快速收敛的算法。重新调整块迭代期望最大化最大似然法 (RBI-EMML) 与 RBI-MART 密切相关，但不是 BI-SUMMA 的特例。
The recently presented sequential unconstrained minimization algorithm, SUMMA, is extended to provide a framework for the derivation of block-iterative, or partial-gradient, optimization methods. This block-iterative SUMMA (BI-SUMMA) includes, and is motivated by, block-iterative versions of the algebraic reconstruction technique (ART) and its multiplicative variant, the MART. The BI-SUMMA approach is designed to provide computationally tractable and quickly convergent algorithms. The rescaled block-iterative expectation maximization maximum likelihood method (RBI-EMML) is closely related to the RBI-MART, but is not a particular case of BI-SUMMA.