稀疏编码自编码表达
From Ufldl
(→Good initialization of s[良好的s初始值]) |
(→Sparse coding in practice) |
||
| Line 256: | Line 256: | ||
[原文] | [原文] | ||
| - | == Sparse coding in practice == | + | == Sparse coding in practice[稀疏编码实践] == |
As suggested in the earlier sections, while the theory behind sparse coding is quite simple, writing a good implementation that actually works and converges reasonably quickly to good optima requires a bit of finesse. | As suggested in the earlier sections, while the theory behind sparse coding is quite simple, writing a good implementation that actually works and converges reasonably quickly to good optima requires a bit of finesse. | ||
| Line 262: | Line 262: | ||
[初译] | [初译] | ||
| + | 如上所述，稀疏编码背后的理论十分简单，但是要实现一个好的稀疏编码算法需要一定的技巧。一个好的算法不仅能实际工作并且能够合理快速地收敛到最优值。 | ||
[一审] | [一审] | ||
| - | + | 如上所述，稀疏编码背后的理论十分简单，但是要实现一个好的稀疏编码算法需要一定的技巧。一个好的算法不仅能实际工作并且能够合理快速地收敛到最优值。 | |
| + | [原文] | ||
Recall the simple iterative algorithm proposed earlier: | Recall the simple iterative algorithm proposed earlier: | ||
| Line 280: | Line 282: | ||
[初译] | [初译] | ||
| + | 调用之前提到的简单迭代算法： | ||
| + | <ol> | ||
| + | <li>随机初始化<math>A</math> | ||
| + | <li>重复以下步骤直至收敛： | ||
| + | <ol> | ||
| + | <li>根据上一步给定的<math>A</math>，求解能够最小化<math>J(A, s)</math>的<math>s</math> | ||
| + | <li>根据上一步得到的<math>s</math>，求解能够最小化<math>J(A, s)</math>的<math>A</math> | ||
| + | </ol> | ||
| + | </ol> | ||
[一审] | [一审] | ||
| + | |||
| + | 调用之前提到的简单迭代算法： | ||
| + | <ol> | ||
| + | <li>随机初始化<math>A</math> | ||
| + | <li>重复以下步骤直至收敛： | ||
| + | <ol> | ||
| + | <li>根据上一步给定的<math>A</math>，求解能够最小化<math>J(A, s)</math>的<math>s</math> | ||
| + | <li>根据上一步得到的<math>s</math>，求解能够最小化<math>J(A, s)</math>的<math>A</math> | ||
| + | </ol> | ||
| + | </ol> | ||
[原文] | [原文] | ||
| - | |||
It turns out that running this algorithm out of the box will not produce very good results, if any results are produced at all. There are two main tricks to achieve faster and better convergence: | It turns out that running this algorithm out of the box will not produce very good results, if any results are produced at all. There are two main tricks to achieve faster and better convergence: | ||
| Line 294: | Line 314: | ||
[初译] | [初译] | ||
| + | 事实证明，即使有结果产生该算法的开箱即用也不会产生很好的效果。以下是两个快速收敛技巧： | ||
| + | <ol> | ||
| + | <li>将样本处理为“迷你块” | ||
| + | <li>良好的<math>s</math>初始值 | ||
| + | </ol> | ||
[一审] | [一审] | ||
| + | |||
| + | 事实证明，即使有结果产生该算法的开箱即用也不会产生很好的效果。以下是两个快速收敛技巧： | ||
| + | <ol> | ||
| + | <li>将样本处理为“迷你块” | ||
| + | <li>良好的<math>s</math>初始值 | ||
| + | </ol> | ||
[原文] | [原文] | ||
