差分
このページの2つのバージョン間の差分を表示します。
| 両方とも前のリビジョン 前のリビジョン 次のリビジョン | 前のリビジョン 次のリビジョン両方とも次のリビジョン | ||
| playground:playground [2021/08/22 14:25] – Hideaki IIDUKA | playground:playground [2021/08/22 14:32] – Hideaki IIDUKA | ||
|---|---|---|---|
| 行 5: | 行 5: | ||
| | | Constant learning rate | Diminishing learning rate | Constant learning rate | Diminishing learning rate | | | | Constant learning rate | Diminishing learning rate | Constant learning rate | Diminishing learning rate | | ||
| | SGD \cite{sca2020}|$\displaystyle{\mathcal{O}\left( \frac{1}{T} \right) + C}$|$\displaystyle{\mathcal{O}\left( \frac{1}{\sqrt{T}} \right)}$|$\displaystyle{\mathcal{O}\left( \frac{1}{n} \right) + C}$| $\displaystyle{\mathcal{O}\left( \frac{1}{\sqrt{n}} \right)}$ | | | SGD \cite{sca2020}|$\displaystyle{\mathcal{O}\left( \frac{1}{T} \right) + C}$|$\displaystyle{\mathcal{O}\left( \frac{1}{\sqrt{T}} \right)}$|$\displaystyle{\mathcal{O}\left( \frac{1}{n} \right) + C}$| $\displaystyle{\mathcal{O}\left( \frac{1}{\sqrt{n}} \right)}$ | | ||
| - | | SGD with SPS \cite{loizou2021}| -- |$\displaystyle{\mathcal{O}\left( \frac{1}{T} \right) + C}$|---------|$\displaystyle{\mathcal{O}\left( \frac{1}{n} \right) + C}$| | + | | SGD with SPS \cite{loizou2021}|---------|$\displaystyle{\mathcal{O}\left( \frac{1}{T} \right) + C}$|---------|$\displaystyle{\mathcal{O}\left( \frac{1}{n} \right) + C}$| |
| + | | Minibatch SGD \cite{chen2020}|---------|$\displaystyle{\mathcal{O}\left( \frac{1}{T} \right) + C}$|---------|$\displaystyle{\mathcal{O}\left( \frac{1}{n} \right) + C}$| | ||
| + | | Adam \cite{adam}|---------|$\displaystyle{\mathcal{O}\left( \frac{1}{\sqrt{T}} \right)}^{(*)}$|---------|---------| | ||
| + | | AMSGrad \cite{reddi2018}|---------|$\displaystyle{\mathcal{O}\left( \sqrt{\frac{1 + \ln T}{T}} \right)}$|---------|---------| | ||
| + | | GWDC \cite{liang2020}|---------|$\displaystyle{\mathcal{O}\left( \frac{1}{\sqrt{T}} \right)}$|---------|---------| | ||
| + | | AMSGWDC \cite{liang2020}|---------|$\displaystyle{\mathcal{O}\left( \frac{1}{\sqrt{T}} \right)}$|---------|---------| | ||
| + | | AMSGrad \cite{chen2019}|---------|$\displaystyle{\mathcal{O}\left( \frac{\ln T}{\sqrt{T}} \right)}$|--------|$\displaystyle{\mathcal{O}\left( \frac{\ln n}{\sqrt{n}} \right)}$| | ||
| - | Minibatch SGD \cite{chen2020} | ||
| - | & --------- | ||
| - | & $\displaystyle{\mathcal{O}\left( \frac{1}{T} \right) + C}$ | ||
| - | & --------- | ||
| - | & $\displaystyle{\mathcal{O}\left( \frac{1}{n} \right) + C}$ \\ \hline\hline | ||
| - | Adam \cite{adam} | ||
| - | & --------- | ||
| - | & $\displaystyle{\mathcal{O}\left( \frac{1}{\sqrt{T}} \right)}^{(*)}$ | ||
| - | |||
| - | & --------- | ||
| - | & --------- \\ | ||
| - | AMSGrad \cite{reddi2018} | ||
| - | & --------- | ||
| - | & $\displaystyle{\mathcal{O}\left( \sqrt{\frac{1 + \ln T}{T}} \right)}$ | ||
| - | |||
| - | & --------- | ||
| - | & --------- \\ | ||
| - | GWDC \cite{liang2020} | ||
| - | & --------- | ||
| - | & $\displaystyle{\mathcal{O}\left( \frac{1}{\sqrt{T}} \right)}$ | ||
| - | & --------- | ||
| - | & --------- \\ | ||
| - | AMSGWDC \cite{liang2020} | ||
| - | & --------- | ||
| - | & $\displaystyle{\mathcal{O}\left( \frac{1}{\sqrt{T}} \right)}$ | ||
| - | & --------- | ||
| - | & --------- \\ | ||
| - | AMSGrad \cite{chen2019} | ||
| - | & --------- | ||
| - | & $\displaystyle{\mathcal{O}\left( \frac{\ln T}{\sqrt{T}} \right)}$ | ||
| - | |||
| - | & --------- | ||
| - | & $\displaystyle{\mathcal{O}\left( \frac{\ln n}{\sqrt{n}} \right)}$ \\ | ||
| AdaBelief \cite{adab} | AdaBelief \cite{adab} | ||
| & --------- | & --------- | ||