Fischer, Lydia: Rejection and online learning with prototype-based classifiers in adaptive metrical spaces. 2016
Inhalt
- Introduction
- Motivation
- Contribution of this Thesis
- Structural Overview of this Thesis
- Publications and Funding Related to this Thesis
- Principles of Rejection
- Prototype-based Classification
- Generalised Learning Vector Quantisation
- Generalised Matrix Learning Vector Quantisation
- Localised Generalised Matrix Learning Vector Quantisation
- Robust Soft Learning Vector Quantisation
- Global Reject Option
- Motivation
- Research Questions
- Certainty Measures
- Experiments for Global Rejection
- Comparison with Probabilistic Approaches
- Gaussian Mixture Model and its Certainty Measure
- Experiments
- Conclusion with Respect to Probabilistic Approaches
- Conclusion: Answering the Research Questions
- Local Reject Option
- Motivation
- Research Questions
- Classifiers
- Prototype-based Classifiers
- Basic Decision Trees for Classification
- Support Vector Machine for Classification
- Local Rejection
- Optimal Choices of Rejection Thresholds
- Extended Pareto Front
- Optimal Global Rejection
- Optimal Local Rejection
- Formulation as Multiple Choice Knapsack Problem
- Local Threshold Adaptation by Dynamic Programming
- Local Threshold Adaptation by an Efficient Greedy Strategy
- Experiments for Local Rejection
- Data Sets
- Dynamic Programming versus Greedy Optimisation
- Experiments on Artificial Data
- Experiments on Benchmarks
- Medical Application – The Adrenal Tumours Data
- Conclusion: Answering the Research Questions
- Incremental Online Learning Vector Quantisation
- Motivation
- Research Questions
- Related Work
- Incremental Online Learning Vector Quantisation
- Experiments
- Influence of Parameters for Incremental Learning
- Compatibility with Metric Learning
- Comparative Evaluation
- Conclusion: Answering the Research Questions
- Combined Offline and Online Learning
- Motivation
- Description of the Scenario
- Research Questions
- Related Work
- Combining Offline and Online Learning
- Experiments on Artificial and Benchmark Data
- Summary of the Main Findings
- Online Metric Learning for an Adaptation to Confidence Drift
- Conclusion: Answering the Research Questions
- Application on Road Terrain Detection
- Motivation
- Research Questions
- Road Terrain Detection – Related Work
- The Road Terrain Detection System
- The Scenario
- Experimental Studies
- Conclusion: Answering the Research Questions
- Conclusion
- Appendix
- Bibliography