/Lion-vs-Adam

Lion and Adam optimization comparison

Primary LanguageJupyter Notebook

🦁 Lion (EvoLved Sign Momentum) vs Adam(W)

Lion, EvoLved Sign Momentum , Google Brain发现的新优化器,据称比Adam(W)更好。

对Lion与Adam(W)及其他接近的优化器进行了二维优化的可视化对比,发现:

  • Lion优化器趋向于在大范围进行搜索解空间以后再进行小范围搜索
  • Adam(W) 优化器趋向于直接使用小范围局部的梯度直接走向局部最小值,此外SGD也有类似的表现
  • SIGNUM 优化器则与Lion相近,但是由于其没有衰减系数,无法快速的锁定大致范围,而在大范围内跳跃。
  • Lion优化器必须搭配学习率衰减,适配的学习率衰减策略应该在大学习率时有效的跳过局部最小值,并快速锁定全局最小值所在范围,然后使用小学习率收敛至全局最小值。

With lr decay

Lion

Adam

SGD

SIGNUM

Without lr decay

Lion

Adam

SGD

SIGNUM

Citations

@misc{https://doi.org/10.48550/arxiv.2302.06675,
    url     = {https://arxiv.org/abs/2302.06675},
    author  = {Chen, Xiangning and Liang, Chen and Huang, Da and Real, Esteban and Wang, Kaiyuan and Liu, Yao and Pham, Hieu and Dong, Xuanyi and Luong, Thang and Hsieh, Cho-Jui and Lu, Yifeng and Le, Quoc V.},
    title   = {Symbolic Discovery of Optimization Algorithms},
    publisher = {arXiv},
    year = {2023}
}