0.25
0.5
0.75
1.25
1.5
1.75
2
On manifolds and autoencoders
Published on Sep 13, 201513568 Views
Related categories
Chapter list
On manifolds and autoencoders00:00
Plan00:16
An unsupervised learning task: dimensionality reduction - 101:30
An unsupervised learning task: dimensionality reduction - 202:45
A classic algorithm Principal Component Analysis - 103:23
A classic algorithm Principal Component Analysis - 205:42
Lower-dimensional manifolds embedded in high dimensional space06:18
The manifold hypothesis07:34
The curse of dimensionality - 108:39
The curse of dimensionality - 210:01
The curse of dimensionality - 310:15
Manifold follows naturally from continuous underlying factors11:51
Modeling local tangent spaces20:52
Non-parametric density estimation21:12
Non-local manifold Parzen windows - 124:40
Non-local manifold Parzen windows - 225:15
Manifold learning is a rich subfield26:11
Neighborhood-based training! - 126:53
Neighborhood-based training! - 227:13
Neighborhood-based training! - 327:19
PART II27:35
Multi-Layer Perceptron (MLP)27:56
Autoencoders: MLPs used for «unsupervised» representation learning30:10
Auto-Encoders (AE) for learning representations31:57
conection between Linear auto-encoders and PCA37:16
similarity between Auto-encoders and RBM38:22
Greedy Layer-Wise Pre-training with RBMs - 141:17
Greedy Layer-Wise Pre-training with RBMs - 242:32
Supervised fine-tuning44:48
Supervised Fine-Tuning is Important51:30
Basic auto-encoders not as good feature learners as RBMs...53:31
Denoising auto-encoders: motivation56:42
Denoising auto-encoder (DAE) - 158:04
Denoising auto-encoder (DAE) - 258:37
Denoising auto-encoder (DAE) - 359:17
Denoising auto-encoder (DAE) - 459:31
Denoising auto-encoder (DAE) - 501:00:07
Denoising auto-encoder (DAE) - 601:02:08
Learned filters - 101:02:10
Learned filters - 201:03:20
Denoising auto-encoders: manifold interpretation01:05:31
Stacked Denoising Auto-Encoders (SDAE)01:08:49
Encouraging representation to be insensitive to corruption - 101:13:11
Encouraging representation to be insensitive to corruption - 201:14:30
From stochastic to analytic penalty01:15:31
Contractive Auto-Encoder (CAE) - 101:17:35
Contractive Auto-Encoder (CAE) - 201:18:22
Computational considerations CAE for a simple encoder layer01:18:24
Higher order Contractive Auto-Encoder (CAE+H) - 101:21:52
Higher order Contractive Auto-Encoder (CAE+H) - 201:22:57
Learned filters01:23:03
CAE must capture manifold directions - 101:23:23
CAE must capture manifold directions - 201:24:03
CAE must capture manifold directions - 301:24:14
CAE must capture manifold directions - 401:24:38
CAE must capture manifold directions - 501:25:29
CAE must capture manifold directions - 601:25:38
Learned tangent space01:26:02
SVD01:28:57
Learned tangents CIFAR-10 - 101:30:25
Learned tangents CIFAR-10 - 201:30:31
How to leverage the learned tangents - 101:31:15
How to leverage the learned tangents - 201:32:32