Learning from Weakly Labeled Data thumbnail
Pause
Mute
Subtitles
Playback speed
0.25
0.5
0.75
1
1.25
1.5
1.75
2
Full screen

Learning from Weakly Labeled Data

Published on Aug 26, 20135365 Views

In many machine learning problems, the labels of the training examples are incomplete. These include, for example, (i) semi-supervised learning where labels are partially known; (ii) multi-instance l

Related categories

Chapter list

Learning from Weakly Labeled Data00:00
Outline00:09
Introduction00:35
Semi-Supervised Learning (SSL)01:14
Multiple Instance Learning (MIL)02:13
Weak Label Information03:45
Clustering05:23
Weak-Label Learning06:20
Existing Algorithms06:46
WellSVM (WEakly LabeLed SVM)08:28
Large-Margin Weak-Label Learning10:59
Large Margin Classifiers13:22
Relax15:08
Tightest Convex Relaxation17:24
How to Solve?19:30
Cutting Plane Algorithm by Label Generation21:15
Properties22:55
Semi-Supervised Learning - 124:45
Semi-Supervised Learning - 226:10
Cutting Plane Algorithm27:04
Issue 1: How to obtain α?27:24
Multiple Label-Kernel Learning29:33
Issue 2: Finding a Violated Label Assignment31:15
Simple Method to Find a Violated Label Assignment33:10
WellSVM for Semi-Supervised Learning35:07
Experiments35:24
Accuracies (5% labeled examples)37:07
CPU Time37:12
Number of WellSVM Iterations37:30
Larger Data Sets37:48
Comparison with Other SSL Benchmarks in the Literature38:27
Comparison with SDP-based Benchmarks38:54
Multiple Instance Learning39:03
Multiple Instance SVM39:46
Optimization Problem41:13
Convex Relaxation41:53
Cutting Plane: Step 142:23
Cutting Plane: Step 242:56
Experiment: CBIR44:11
Multiple Instance Learning for Locating ROIs45:11
Location the Region of Interest (ROI)45:46
Maximum Margin Clustering46:21
Cutting Plane46:52
Experiments47:17
Clustering Accuracies47:30
CPU Time47:37
Large-Scale Experiments (Linear Kernel)48:00
Conclusion48:04
References48:59