(Downloads - 0)
For more info about our services contact : help@bestpfe.com
Table of contents
1 Introduction
1.1 Large Scale Machine Learning
1.1.1 Machine Learning
1.1.2 Towards Large Scale Applications
1.1.3 Online Learning
1.1.4 Scope of this Thesis
1.2 New Efficient Algorithms for Support Vector Machines
1.2.1 A New Generation of Online SVM Dual Solvers
1.2.2 A Carefully Designed Second-Order SGD
1.2.3 A Learning Method for Ambiguously Supervised SVMs
1.2.4 Careful Implementations
1.3 Outline of the Thesis
2 Support Vector Machines
2.1 Kernel Classifiers
2.1.1 Support Vector Machines
2.1.2 Solving SVMs with SMO
2.1.3 Online Kernel Classifiers
2.1.4 Solving Linear SVMs
2.2 SVMs for Structured Output Prediction
2.2.1 SVM Formulation
2.2.2 Batch Structured Output Solvers
2.2.3 Online Learning for Structured Outputs
2.3 Summary
3 Efficient Learning of Linear SVMs with Stochastic Gradient Descent
3.1 Stochastic Gradient Descent
3.1.1 Analysis
3.1.2 Scheduling Stochastic Updates to Exploit Sparsity
3.1.3 Implementation
3.2 SGD-QN: A Careful Diagonal Quasi-Newton SGD
3.2.1 Rescaling Matrices
3.2.2 SGD-QN
3.2.3 Experiments
3.3 Summary
4 Large-Scale SVMs for Binary Classification
4.1 The Huller: an Efficient Online Kernel Algorithm
4.1.1 Geometrical Formulation of SVMs
4.1.2 The Huller Algorithm
4.1.3 Experiments
4.1.4 Discussion
4.2 Online LaSVM
4.2.1 Building Blocks
4.2.2 Scheduling
4.2.3 Convergence and Complexity
4.2.4 Implementation Details
4.2.5 Experiments
4.3 Active Selection of Training Examples
4.3.1 Example Selection Strategies
4.3.2 Experiments on Example Selection for Online SVMs
4.3.3 Discussion
4.4 Tracking Guarantees for Online SVMs
4.4.1 Analysis Setup
4.4.2 Duality Lemma
4.4.3 Algorithms and Analysis
4.4.4 Application to LaSVM
4.5 Summary
5 Large-Scale SVMs for Structured Output Prediction
5.1 Structured Output Prediction with LaRank
5.1.1 Elementary Step
5.1.2 Step Selection Strategies
5.1.3 Scheduling
5.1.4 Stopping
5.1.5 Theoretical Analysis
5.2 Multiclass Classification
5.2.1 Multiclass Factorization
5.2.2 LaRank Implementation for Multiclass Classification
5.2.3 Experiments
5.3 Sequence Labeling
5.3.1 Representation and Inference
5.3.2 Training
5.3.3 LaRank Implementations for Sequence Labeling
5.3.4 Experiments
5.4 Summary
6 Learning SVMs under Ambiguous Supervision
6.1 Online Multiclass SVM with Ambiguous Supervision
6.1.1 Classification with Ambiguous Supervision
6.1.2 Online Algorithm
6.2 Sequential Semantic Parser
6.2.1 The OSPAS Algorithm
6.2.2 Experiments
6.3 Summary
7 Conclusion
7.1 Large Scale Perspectives for SVMs
7.1.1 Impact and Limitations of our Contributions
7.1.2 Further Derivations
7.2 AI Directions
7.2.1 Human Homology
7.2.2 Natural Language Understanding
Bibliography
A Personal Bibliography
B Convex Programming with Witness Families
B.1 Feasible Directions
B.2 Witness Families
B.3 Finite Witness Families
B.4 Stochastic Witness Direction Search
B.5 Approximate Witness Direction Search
B.5.1 Example (SMO)
B.5.2 Example (LaSVM)
B.5.3 Example (LaSVM + Gradient Selection)
B.5.4 Example (LaSVM + Active Selection + Randomized Search)
C Learning to Disambiguate Language Using World Knowledge
C.1 Introduction
C.2 Previous Work
C.3 The Concept Labeling Task
C.4 Learning Algorithm
C.5 A Simulation Environment
C.5.1 Universe Definition
C.5.2 Simulation Algorithm
C.6 Experiments
C.7 Weakly Labeled Data
C.8 Conclusion
