introduction to chap13 - statistical learning and machine...

73
Introduction to Statistical Learning and Machine Learning Chap13 - Semi-supervised Learning Fudan-SDS Confidential - Do Not Distribute

Upload: others

Post on 16-Aug-2020

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Introduction to Statistical Learning

and Machine Learning

Chap13 - Semi-supervised Learning

Fudan-SDS Confidential - Do Not Distribute

Page 2: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Chap12 - Semi-supervised Learning

Fudan-SDS Confidential - Do Not Distribute

Page 3: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleBooks

Page 4: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitle

Page 5: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Why bother?

Page 6: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitlewhile it may be dangerous to obtain labelled data.

Page 7: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitlewhile it may be dangerous to obtain labelled data.

Page 8: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleExamples of hard-to-get labels

Page 9: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleNotations

Page 10: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Learning paradigms

Page 11: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleWhy the name

Page 12: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleWhy the name

We will mainly discuss semi-supervised classification.

Page 13: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Semi-supervised Learning

Intuition understanding

Fudan-SDS Confidential - Do Not Distribute

Page 14: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleHow semi-supervised learning is helpful?

Page 15: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleHow semi-supervised learning is helpful

Page 16: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleHow semi-supervised learning is helpful?

Page 17: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleTwo moons

There are 2 labeled points (the triangle and the cross) and 100 unlabeled points. The global optimum of S3VM correctly identifies the decision boundary (black line).

This figure comes from Chapelle et al. “Optimization Techniques for Semi-Supervised Support Vector Machines”, JMLR 2007.

Page 18: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleDoes unlabeled data always help?

Page 19: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleDoes unlabeled data always help?

Unfortunately, this is not the case, yet.

Page 20: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleDoes unlabeled data always help?

Unfortunately, this is not the case, yet.

Not until recently, “Safe Semi-supervised learning”, Yu-Feng Li and Zhi-Hua Zhou. Towards making unlabeled data never hurt. IEEE Transactions on Pattern Analysis and Machine Intelligence 2014.

Page 21: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleAssumption and Intuitions

• Points in the same cluster likely to be of the same class; • Decision line should lie in a low-density region.

Guess the name?

Page 22: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleAssumption and Intuitions

• Points in the same cluster likely to be of the same class; • Decision line should lie in a low-density region.

Guess the name?

Cluster Assumption

Page 23: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Assumption and Intuitions

• The data lie approximately on a manifold of much lower dimension than the input space. • Swiss Roll

Guess the name?

Page 24: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Assumption and Intuitions

• The data lie approximately on a manifold of much lower dimension than the input space. • Swiss Roll

Guess the name?

Manifold Assumption

Page 25: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleAssumption and Intuitions

If two points x1, x2 in a high-density region are close, then so should be the corresponding outputs y1, y2.

Guess the name?

Points which are close to each other are more likely to share a label. This is also generally assumed in supervised learning and yields a preference for geometrically simple decision boundaries. In the case of semi-supervised learning, the smoothness assumption additionally yields a preference for decision boundaries in low-density regions, so that there are fewer points close to each other but in different classes.

Page 26: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleAssumption and Intuitions

If two points x1, x2 in a high-density region are close, then so should be the corresponding outputs y1, y2.

Guess the name?

Smoothness AssumptionPoints which are close to each other are more likely to share a label. This is also generally assumed in supervised learning and yields a preference for geometrically simple decision boundaries. In the case of semi-supervised learning, the smoothness assumption additionally yields a preference for decision boundaries in low-density regions, so that there are fewer points close to each other but in different classes.

Page 27: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Semi-supervised Learning

Key methods of SSL

Fudan-SDS Confidential - Do Not Distribute

Page 28: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleMethods and Algorithms of SSL

• Self-training • Co-training -Multi-view Algorithm • Graph-based SVM • Semi-supervised SVM (S3VM)

Page 29: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Self-training

Fudan-SDS Confidential - Do Not Distribute

Page 30: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleSelf-Training algorithm

Page 31: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Variations in self-training

Page 32: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Self-training example: image categorisation

Page 33: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Self-training example: image categorisation

Page 34: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

The bag-of-word representation of images

Page 35: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Self-training example: image categorization

Page 36: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Self-training example: image categorization

Page 37: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Comments on Self-trainingAdvantages of self-training

Disadvantages of self-training

Page 38: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Co-training one type of Multi-view algorithm

Fudan-SDS Confidential - Do Not Distribute

Page 39: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Co-training

Page 40: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Feature split

Page 41: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Co-training assumptions

Page 42: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Co-training algorithm

Page 43: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitlePros and cons of co-training

Page 44: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Variants of co-training

Page 45: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Graph-based semi-supervised

learning

Fudan-SDS Confidential - Do Not Distribute

Page 46: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Graph-based semi-supervised learning

Page 47: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Graph-based semi-supervised learning

Page 48: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Graph construction

Page 49: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleGraph-based semi-supervised learning

Page 50: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleAdjacency matrix

Page 51: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Incidence matrixThe unoriented incidence matrix (or simply incidence matrix) of an undirected graph is a n × m matrix B, where n and m are the numbers of vertices and edges respectively, such that Bi,j = 1 if the vertex vi and edge ej are incident and 0 otherwise.

For example the incidence matrix of the undirected graph shown on the right is a matrix consisting of 4 rows (corresponding to the four vertices, 1-4) and 4 columns (corresponding to the four edges, e1-e4):

Page 52: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleQuestions:Incidence matrix & Adjacency matrix

Page 53: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleQuestions:Incidence matrix & Adjacency matrix

Page 54: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleQuestions:Incidence matrix & Adjacency matrix

Page 55: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

The graph

Page 56: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleLabel Propagation on graphs

Page 57: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleHarmonic function f

Semi-Supervised Learning Using Gaussian Fields and Harmonic Functions. Xiaojin Zhu, Zoubin Ghahramani, John Lafferty. ICML-2003. Awarded the classic paper prize in ICML 2013

Page 58: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleAn electric network interpretation

Page 59: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleA random walk interpretation

Page 60: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Label Propagation

Zhou D.-Y. Learning with Local and Global Consistency, NIPS 2004.

Page 61: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleLabel Propagation

Page 62: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleLabel Propagation

Page 63: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleLabel Propagation

Page 64: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleGraph partition

Page 65: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Min-cuts

In graph theory, a minimum cut of a graph is a cut (a partition of the vertices of a graph into two disjoint subsets that are joined by at least one edge) that is minimal in some sense.

Max-flow min-cut theorem:

The cut in a flow network that separates the source and sink vertices and minimizes the total weight on the edges that are directed from the source side of the cut to the sink side of the cut. As shown in the max-flow min-cut theorem, the weight of this cut equals the maximum amount of flow that can be sent from the source to the sink in the given network.

Karger's Min Cut Algorithm

Page 66: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Min-cutsGraph-cut Normalised Cut

In graph theory, a minimum cut of a graph is a cut (a partition of the vertices of a graph into two disjoint subsets that are joined by at least one edge) that is minimal in some sense.

Max-flow min-cut theorem:

The cut in a flow network that separates the source and sink vertices and minimizes the total weight on the edges that are directed from the source side of the cut to the sink side of the cut. As shown in the max-flow min-cut theorem, the weight of this cut equals the maximum amount of flow that can be sent from the source to the sink in the given network.

Karger's Min Cut Algorithm

Page 67: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Karger's minimum cut algorithm

For his dissertation "Random Sampling in Graph Optimization Problems."

David Karger ACM Doctoral Dissertation Award United States – 1994

Page 68: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Semi-supervised Learning

(Transductive) SVM

Fudan-SDS Confidential - Do Not Distribute

Page 69: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

S3VM [Joachims98]• Suppose we believe target separator goes through low density regions of the space/large margin. • Aim for separator with large margin w.r.t labeled and unlabeled data. (L+U)

+

+

_

_

Labeled data only

+

+

_

_

+

+

_

_

S^3VMSVM

Page 70: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleSemi-supervised SVM

Page 71: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleSemi-supervised SVM

Page 72: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Fudan-SDS Confidential - Do Not Distribute

Optional subtitleSemi-supervised SVM

Page 73: Introduction to Chap13 - Statistical Learning and Machine ...yanweifu.github.io/courses/SLML/chap11_semi... · Fudan-SDS ConÞdential - Do Not Distribute Incidence matrix The unoriented

Appendix

Fudan-SDS Confidential - Do Not Distribute