Explanation-Based Neural Network Learning - Thrun, Sebastian

. .

Résumé :
Lifelong learning addresses situations in which a learner faces a series of different learning tasks providing the opportunity for synergy among them. Explanation-based neural network learning (EBNN) is a machine learning algorithm that transfers knowledge across multiple learning tasks. When faced with a new learning task, EBNN exploits domain knowledge accumulated in previous learning tasks to guide generalization in the new one. As a result, EBNN generalizes more accurately from less data than comparable methods. Explanation-Based Neural Network Learning: A Lifelong Learning Approach describes the basic EBNN paradigm and investigates it in the context of supervised learning, reinforcement learning, robotics, and chess. `The paradigm of lifelong learning - using earlier learned knowledge to improve subsequent learning - is a promising direction for a new generation of machine learning algorithms. Given the need for more accurate learning methods, it is difficult to imagine a future for machine learning that does not include this paradigm.' From the Foreword by Tom M. Mitchell.

Sommaire:
1 Introduction.- 1.1 Motivation.- 1.2 Lifelong Learning.- 1.3 A Simple Complexity Consideration.- 1.4 The EBNN Approach to Lifelong Learning.- 1.5 Overview.- 2 Explanation-Based Neural Network Learning.- 2.1 Inductive Neural Network Learning.- 2.2 Analytical Learning.- 2.3 Why Integrate Induction and Analysis?.- 2.4 The EBNN Learning Algorithm.- 2.5 A Simple Example.- 2.6 The Relation of Neural and Symbolic Explanation-Based Learning.- 2.7 Other Approaches that Combine Induction and Analysis.- 2.8 EBNN and Lifelong Learning.- 3 The Invariance Approach.- 3.1 Introduction.- 3.2 Lifelong Supervised Learning.- 3.3 The Invariance Approach.- 3.4 Example: Learning to Recognize Objects.- 3.5 Alternative Methods.- 3.6 Remarks.- 4 Reinforcement Learning.- 4.1 Learning Control.- 4.2 Lifelong Control Learning.- 4.3 Q-Learning.- 4.4 Generalizing Function Approximators and Q-Learning.- 4.5 Remarks.- 5 Empirical Results.- 5.1 Learning Robot Control.- 5.2 Navigation.- 5.3 Simulation.- 5.4 Approaching and Grasping a Cup.- 5.5 NeuroChess.- 5.6 Remarks.- 6 Discussion.- 6.1 Summary.- 6.2 Open Problems.- 6.3 Related Work.- 6.4 Concluding Remarks.- A An Algorithm for Approximating Values and Slopes with Artificial Neural Networks.- A.1 Definitions.- A.2 Network Forward Propagation.- A.3 Forward Propagation of Auxiliary Gradients.- A.4 Error Functions.- A.5 Minimizing the Value Error.- A.6 Minimizing the Slope Error.- A.7 The Squashing Function and its Derivatives.- A.8 Updating the Network Weights and Biases.- B Proofs of the Theorems.- C Example Chess Games.- C.1 Game 1.- C.2 Game 2.- References.- List of Symbols.