基于重启技术的加速邻近梯度算法

Accelerated Proximal Gradient Method Based on Restart Technique

加速邻近梯度法是在梯度法基础上的一个改进,虽然效率比梯度法有明显提高,但仍存在收敛轨迹出现往回迭代的情况。为了克服该缺点,提出了把重启技术应用在加速邻近梯度法上的方法,并通过数值例子进行了比对,证明了该技术的有效性。

The accelerated proximal gradient method is upgraded from gradient method. Although the efficiency of the accelerated proximal gradient method is apparently better than that of the gradient method,the trajectory of the accelerated proximal gradient method may oscillate. To overcome this drawback,restart techniques are applied to the accelerated proximal gradient method. Numerical examples show that the techniques are useful.