Outline Provisional

• Introduction

1. A general model of learning agent
1. The learning element
2. The performance element
2. The learning paradigms
1. The subsymbolic paradigm
2. The symbolic paradigm
3. The learning strategies
1. Inductive learning
2. Deductive learning
3. Abductive learning
4. Multistrategy learning
4. The representation problem
1. Feature vectors and attribute-based representations
2. Structural representations
5. Machine learning issues

• Inductive learning for attribute-based representations

1. Statistical methods
1. Parametric methods
2. Non-parametric methods
3. Relevance for document analysis and recognition
2. Assessing the performance of an inductive learning algorithm
3. Decision trees
1. Main problems in decision tree induction
2. The decision tree learning algorithm
3. Handling continuous attributes
4. Measures for attribute selection
5. Pruning to avoid overfitting the data
6. From decision trees to rules
7. Case study: C4.5
8. Further issues in decision tree learning
9. Incremental learning of decision trees
10. Case study: ITI
11. Relevance for document analysis and recognition

• Inductive learning for structural representations

1. Concept learning as search
1. The instance space and the hypotheses space
2. General-to-specific ordering of hypotheses
3. Generalization
4. Specialization
5. Sequential covering algorithms
2. Learning first-order rules
1. First-order Horn clauses
2. Mapping clauses to graphs and viceversa
3. Generalization
4. Specialization
5. Top-down and bottom-up learning algorithms
6. Case study: FOIL
7. Using prior knowledge in learning
8. Case study: FOCL
9. Relevance for document analysis and recognition

• Learning in document analysis and recognition: open problems and research directions.