Explanation-Based Neural Network Learning

Picture a learner who must tackle a sequence of distinct problems, where what it picks up in one can help with the next. That idea is the core of lifelong learning, and this book examines Explanation-Based Neural Network Learning (EBNN), a machine learning algorithm built to carry knowledge from one task to another. When a new task arrives, EBNN draws on the domain knowledge gathered from earlier tasks to guide how it generalizes, which lets it reach more accurate generalizations from less training data than comparable methods.

The book, subtitled A Lifelong Learning Approach, sets out the basic EBNN paradigm and tests it in several settings: supervised learning, reinforcement learning, robotics, and chess.

The foreword by Tom M. Mitchell argues that reusing earlier knowledge to improve later learning is a direction the next generation of machine learning algorithms will need. In his view, with the demand for more accurate learning methods growing, the future of machine learning is hard to picture without this approach.