Lifelong learning and inductive bias

Effective learning from finite data requires assumptions about the data source, a notion referred to as inductive bias in the Machine Learning literature. A fundamental question pertains to the source of a ‘good’ inductive bias. One natural way to form such a bias is through lifelong learning, where an agent continually interacts with the world through a sequence of tasks, aiming to improve its performance on future tasks based on the tasks it has seen, and solved, so far. We review the problem of lifelong learning within Machine Learning, provide a basic theoretical formulation, and describe some theoretically motivated algorithms that demonstrate the feasibility of inductive bias formation through transfer of knowledge from previous tasks.