Metalearning : Applications to Automated Machine Learning and Data Mining.

Format:

Book

Author/Creator:

Brazdil, Pavel.

Contributor:

van Rijn, Jan N.

Soares, Carlos.

Vanschoren, Joaquin.

Series:

Cognitive Technologies

Language:

English

Physical Description:

1 online resource (349 pages)

Edition:

2nd ed.

Place of Publication:

Cham : Springer International Publishing AG, 2022.

Language Note:

English

Summary:

This open access book as one of the fastest-growing areas of research in machine learning, metalearning studies principled methods to obtain efficient models and solutions by adapting machine learning and data mining processes. This adaptation usually exploits information from past experience on other tasks and the adaptive processes can involve machine learning approaches. As a related area to metalearning and a hot topic currently, automated machine learning (AutoML) is concerned with automating the machine learning processes. Metalearning and AutoML can help AI learn to control the application of different learning methods and acquire new solutions faster without unnecessary interventions from the user. This book offers a comprehensive and thorough introduction to almost all aspects of metalearning and AutoML, covering the basic concepts and architecture, evaluation, datasets, hyperparameter optimization, ensembles and workflows, and also how this knowledge can be used to select, combine, compose, adapt and configure both algorithms and models to yield faster and better solutions to data mining and data science problems. It can thus help developers to develop systems that can improve themselves through experience. This book is a substantial update of the first edition published in 2009. It includes 18 chapters, more than twice as much as the previous version. This enabled the authors to cover the most relevant topics in more depth and incorporate the overview of recent research in the respective area. The book will be of interest to researchers and graduate students in the areas of machine learning, data mining, data science and artificial intelligence. ; Metalearning is the study of principled methods that exploit metaknowledge to obtain efficient models and solutions by adapting machine learning and data mining processes. While the variety of machine learning and data mining techniques now available can, in principle, provide good model solutions, a methodology is still needed to guide the search for the most appropriate model in an efficient way. Metalearning provides one such methodology that allows systems to become more effective through experience. This book discusses several approaches to obtaining knowledge concerning the performance of machine learning and data mining algorithms. It shows how this knowledge can be reused to select, combine, compose and adapt both algorithms and models to yield faster, more effective solutions to data mining problems. It can thus help developers improve their algorithms and also develop learning systems that can improve themselves. The book will be of interest to researchers and graduate students in the areas of machine learning, data mining and artificial intelligence.

Contents:

Intro

Preface

Contents

Part I Basic Concepts and Architecture

1 Introduction

1.1 Organization of the Book

1.2 Basic Concepts and Architecture (Part I)

1.3 Advanced Techniques and Methods (Part II)

1.4 Repositories of Experimental Results (Part III)

References

2 Metalearning Approaches for Algorithm Selection I (Exploiting Rankings)

2.1 Introduction

2.2 Different Forms of Recommendation

2.3 Ranking Models for Algorithm Selection

2.4 Using a Combined Measure of Accuracy and Runtime

2.5 Extensions and Other Approaches

3 Evaluating Recommendations of Metalearning/AutoML Systems

3.1 Introduction

3.2 Methodology for Evaluating Base-Level Algorithms

3.3 Normalization of Performance for Base-Level Algorithms

3.4 Methodology for Evaluating Metalearning and AutoML Systems

3.5 Evaluating Recommendations by Correlation

3.6 Evaluating the Effects of Recommendations

3.7 Some Useful Measures

4 Dataset Characteristics (Metafeatures)

4.1 Introduction

4.2 Data Characterization Used in Classification Tasks

4.3 Data Characterization Used in Regression Tasks

4.4 Data Characterization Used in Time Series Tasks

4.5 Data Characterization Used in Clustering Tasks

4.6 Deriving New Features from the Basic Set

4.7 Selection of Metafeatures

4.8 Algorithm-Specific Characterization and Representation Issues

4.9 Establishing Similarity Between Datasets

5 Metalearning Approaches for Algorithm Selection II

5.1 Introduction

5.2 Using Regression Models in Metalearning Systems

5.3 Using Classification at Meta-level for the Prediction of Applicability

5.4 Methods Based on Pairwise Comparisons

5.5 Pairwise Approach for a Set of Algorithms

5.6 Iterative Approach of Conducting Pairwise Tests

5.7 Using ART Trees and Forests.

5.8 Active Testing

5.9 Non-propositional Approaches

6 Metalearning for Hyperparameter Optimization

6.1 Introduction

6.2 Basic Hyperparameter Optimization Methods

6.3 Bayesian Optimization

6.4 Metalearning for Hyperparameter Optimization

6.5 Concluding Remarks

7 Automating Workflow/Pipeline Design

7.1 Introduction

7.2 Constraining the Search in Automatic Workflow Design

7.3 Strategies Used in Workflow Design

7.4 Exploiting Rankings of Successful Plans (Workflows)

Part II Advanced Techniques and Methods

8 Setting Up Configuration Spaces and Experiments

8.1 Introduction

8.2 Types of Configuration Spaces

8.3 Adequacy of Configuration Spaces for Given Tasks

8.4 Hyperparameter Importance and Marginal Contribution

8.5 Reducing Configuration Spaces

8.6 Configuration Spaces in Symbolic Learning

8.7 Which Datasets Are Needed?

8.8 Complete versus Incomplete Metadata

8.9 Exploiting Strategies from Multi-armed Bandits to Schedule Experiments

8.10 Discussion

9 Combining Base-Learners into Ensembles

9.1 Introduction

9.2 Bagging and Boosting

9.3 Stacking and Cascade Generalization

9.4 Cascading and Delegating

9.5 Arbitrating

9.6 Meta-decision Trees

9.7 Discussion

10 Metalearning in Ensemble Methods

10.1 Introduction

10.2 Basic Characteristics of Ensemble Systems

10.3 Selection-Based Approaches for Ensemble Generation

10.4 Ensemble Learning (per Dataset)

10.5 Dynamic Selection of Models (per Instance)

10.6 Generation of Hierarchical Ensembles

10.7 Conclusions and Future Research

11 Algorithm Recommendation for Data Streams

11.1 Introduction

11.2 Metafeature-Based Approaches

11.3 Data Stream Ensembles

11.4 Recurring Meta-level Models.

11.5 Challenges for Future Research

12 Transfer of Knowledge Across Tasks

12.1 Introduction

12.2 Background, Terminology, and Notation

12.3 Learning Architectures in Transfer Learning

12.4 A Theoretical Framework

13 Metalearning for Deep Neural Networks

13.1 Introduction

13.2 Background and Notation

13.3 Metric-Based Metalearning

13.4 Model-Based Metalearning

13.5 Optimization-Based Metalearning

13.6 Discussion and Outlook

14 Automating Data Science

14.1 Introduction

14.2 Defining the Current Problem/Task

14.3 Identifying the Task Domain and Knowledge

14.4 Obtaining the Data

14.5 Automating Data Preprocessing and Transformation

14.6 Automating Model and Report Generation

15 Automating the Design of Complex Systems

15.1 Introduction

15.2 Exploiting a Richer Set of Operators

15.3 Changing the Granularity by Introducing New Concepts

15.4 Reusing New Concepts in Further Learning

15.5 Iterative Learning

15.6 Learning to Solve Interdependent Tasks

Part III Organizing and Exploiting Metadata

16 Metadata Repositories

16.1 Introduction

16.2 Organizing the World Machine Learning Information

16.3 OpenML

17 Learning from Metadata in Repositories

17.1 Introduction

17.2 Performance Analysis of Algorithms per Dataset

17.3 Performance Analysis of Algorithms across Datasets

17.4 Effect of Specific Data/Workflow Characteristics on Performance

17.5 Summary

18 Concluding Remarks

18.1 Introduction

18.2 Form of Metaknowledge Used in Different Approaches

18.3 Future Challenges

Index.

Notes:

Description based on publisher supplied metadata and other sources.

ISBN:

3-030-67024-4

OCLC:

1301265010