Learning Similarity Metrics with Invariance Properties

author: Yann LeCun, New York University

Categories

Top: Computer Science: Machine Learning
Top: Computer Science: Machine Learning: Structured Output

You might be experiencing some problems with Your Video player.

*Slides*
0:03	Learning Similarity Metrics with Invariances
0:21	Learning an Invariant Dissimilarity Metric
2:36	Dissimilarity Metric for Face Recognition
3:54	Siamese Architecture
4:50	Dissimilarity Metric vs Traditional Classification
5:10	Trainable Metric vs Other Dimensionality Reduction Methods
7:17	Trainable Metrics vs handcrafted invariances
7:49	Siamese Architecture for Comparing TimeSeries Data
8:41	1D Convolutional Net (TDNN)
8:44	Examples
8:55	Siamese Architecture
9:09	Probabilistic Training: Maximum Likelihood
11:07	Solution?
12:23	Another Loss Function
15:31	Loss Function
16:58	Examples of Loss Functions
17:26	Loss Function: SquareExponential
17:37	Face Verification datasets: AT&T, FERET, and AR/Purdue
19:04	Face Verification datasets: AT&T, FERET, and AR/Purdue
19:25	Face Verification datasets: AT&T, FERET, and AR/Purdue
19:57	Face Verification dataset: AR/Purdue
20:04	Preprocessing
20:17	Centering with a Gaussianblurred face template
20:36	Alternated Convolutions and Subsampling
20:41	Architecture for the Mapping Function Gw(X)
21:02	Internal state for genuine and impostor pairs
22:25	Gaussian Face Model in the output space
23:05	Dataset for Verification - Verification Results
23:50	Classification Examples
25:07	lInternal State
25:29	DrLim: Dimensionality; Reduction by Learning an Invariant Mapping
32:10	“Traditional” Manifold Learning
33:17	Learning a FUNCTION from input to output
33:24	Learning an INVARIANT FUNCTION from input to output
34:32	Learning Invariant Manifolds with EBMs
35:29	Step 1: Building a Neighborhood Graph
36:12	Step 2: Pick a Parameterized Family of Function
36:17	Step 3: Pick a Loss function and Minimize it w.r.t. W
36:31	Architecture
36:35	Architecture and loss function
37:08	Loss function
37:34	Mechanical Analogy
37:54	MNIST Dataset
38:01	MNIST Handwritten Digits. Sanity Check
39:01	Architecture of the Gw(X) Function:
39:04	Alternated Convolutions and Subsampling
39:06	Learning a mapping that is invariant to shifts
39:27	Simple Experiment with Shifted MNIST
39:42	Shifted MNIST: LLE Result
41:19	Shifted MNIST: Injecting Prior Knowledge
42:35	Discovering the Viewpoint Manifold
44:14	Generic Object Detection and Recognition with Invariance to Pose and Illumination
44:17	Data Collection, Sample Generation
44:19	NORB Dataset: LLE
45:22	Automatic Discovery of the Viewpoint Manifold with Invariant to Illumination
46:56	NORB Dataset: Learned Hidden Units

Lecture rating

People found this lecture:
Worth seeing
because it is:
Valuable and informative
Well presented
Easily understandable
Acceptably recorded
You need to login to cast your vote.

Report a problem or upload files

If you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.