Find centralized, trusted content and collaborate around the technologies you use most.
Teams
Q&A for work
Connect and share knowledge within a single location that is structured and easy to search.
我已经用 SGD 训练了 CNN,而且训练得很好。但是,一旦我用 Adam 求解器训练模型,100k几乎在迭代之后,它就开始增加损失值。你能帮我解释一下吗?
100k
下图显示solver.prototxt:
solver.prototxt
momentum: 0.99 momentum2: 0.999 #+ test_interval: 1000 test_iter: 40 weight_decay: 0.0005 base_lr: 0.0001