最优化算法
特点:具有超线性收敛速度,只需要计算梯度,避免计算二阶导数
算法步骤
(step0:)
给定初始值(x_0),容许误差(epsilon)
(step1:)
计算梯度(g_k=
abla f(x_k)),if (norm(g_k)<=epsilon), (break;)
输出当前值(x_k)
else (to step2)
(step2:)
[egin{cases} d_k=-g_k, & ext {$k$=0} d_k=-g_k+eta_{k-1}d_{k-1}, & ext {$k$>=1} end{cases}
]